mirrors
/
PaddleClas
mirror of https://github.com/PaddlePaddle/PaddleClas.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #987 from littletomatodonkey/dev/fix_dev_dim 

fix class_dim
pull/988/head
Bin Lu 2021-06-29 18:20:01 +08:00 committed by GitHub
parent b414737545 4d496f787b
commit e744db271b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
36 changed files with 1332 additions and 761 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
ppcls/arch/backbone/model_zoo/alexnet.py
View File

@ -23,10 +23,14 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"AlexNet": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/AlexNet_pretrained.pdparams"}
MODEL_URLS = {
"AlexNet":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/AlexNet_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
class ConvPoolLayer(nn.Layer):
def __init__(self,
input_channels,
@ -64,7 +68,7 @@ class ConvPoolLayer(nn.Layer):
class AlexNetDY(nn.Layer):
def __init__(self, class_dim=1000):
def __init__(self, class_num=1000):
super(AlexNetDY, self).__init__()
stdv = 1.0 / math.sqrt(3 * 11 * 11)
@ -119,7 +123,7 @@ class AlexNetDY(nn.Layer):
name="fc7_offset", initializer=Uniform(-stdv, stdv)))
self._fc8 = Linear(
in_features=4096,
out_features=class_dim,
out_features=class_num,
weight_attr=ParamAttr(
name="fc8_weights", initializer=Uniform(-stdv, stdv)),
bias_attr=ParamAttr(
@ -143,6 +147,7 @@ class AlexNetDY(nn.Layer):
x = self._fc8(x)
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -155,7 +160,9 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
"pretrained type is not available. Please use `string` or `boolean` type."
)
def AlexNet(pretrained=False, use_ssld=False, **kwargs):
model = AlexNetDY(**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["AlexNet"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["AlexNet"], use_ssld=use_ssld)
return model

17
ppcls/arch/backbone/model_zoo/darknet.py
View File

@ -23,10 +23,14 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"DarkNet53": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DarkNet53_pretrained.pdparams"}
MODEL_URLS = {
"DarkNet53":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DarkNet53_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
class ConvBNLayer(nn.Layer):
def __init__(self,
input_channels,
@ -77,7 +81,7 @@ class BasicBlock(nn.Layer):
class DarkNet(nn.Layer):
def __init__(self, class_dim=1000):
def __init__(self, class_num=1000):
super(DarkNet, self).__init__()
self.stages = [1, 2, 8, 8, 4]
@ -126,7 +130,7 @@ class DarkNet(nn.Layer):
stdv = 1.0 / math.sqrt(1024.0)
self._out = Linear(
1024,
class_dim,
class_num,
weight_attr=ParamAttr(
name="fc_weights", initializer=Uniform(-stdv, stdv)),
bias_attr=ParamAttr(name="fc_offset"))
@ -172,6 +176,7 @@ class DarkNet(nn.Layer):
x = self._out(x)
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -183,8 +188,10 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def DarkNet53(pretrained=False, use_ssld=False, **kwargs):
model = DarkNet(**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DarkNet53"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["DarkNet53"], use_ssld=use_ssld)
return model

39
ppcls/arch/backbone/model_zoo/densenet.py
View File

@ -28,12 +28,18 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"DenseNet121": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet121_pretrained.pdparams",
"DenseNet161": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet161_pretrained.pdparams",
"DenseNet169": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet169_pretrained.pdparams",
"DenseNet201": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet201_pretrained.pdparams",
"DenseNet264": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet264_pretrained.pdparams",
}
MODEL_URLS = {
"DenseNet121":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet121_pretrained.pdparams",
"DenseNet161":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet161_pretrained.pdparams",
"DenseNet169":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet169_pretrained.pdparams",
"DenseNet201":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet201_pretrained.pdparams",
"DenseNet264":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DenseNet264_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -196,7 +202,7 @@ class ConvBNLayer(nn.Layer):
class DenseNet(nn.Layer):
def __init__(self, layers=60, bn_size=4, dropout=0, class_dim=1000):
def __init__(self, layers=60, bn_size=4, dropout=0, class_num=1000):
super(DenseNet, self).__init__()
supported_layers = [121, 161, 169, 201, 264]
@ -269,7 +275,7 @@ class DenseNet(nn.Layer):
self.out = Linear(
num_features,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_weights"),
bias_attr=ParamAttr(name="fc_offset"))
@ -289,6 +295,7 @@ class DenseNet(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -301,31 +308,37 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
"pretrained type is not available. Please use `string` or `boolean` type."
)
def DenseNet121(pretrained=False, use_ssld=False, **kwargs):
model = DenseNet(layers=121, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DenseNet121"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["DenseNet121"], use_ssld=use_ssld)
return model
def DenseNet161(pretrained=False, use_ssld=False, **kwargs):
model = DenseNet(layers=161, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DenseNet161"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["DenseNet161"], use_ssld=use_ssld)
return model
def DenseNet169(pretrained=False, use_ssld=False, **kwargs):
model = DenseNet(layers=169, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DenseNet169"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["DenseNet169"], use_ssld=use_ssld)
return model
def DenseNet201(pretrained=False, use_ssld=False, **kwargs):
model = DenseNet(layers=201, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DenseNet201"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["DenseNet201"], use_ssld=use_ssld)
return model
def DenseNet264(pretrained=False, use_ssld=False, **kwargs):
model = DenseNet(layers=264, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DenseNet264"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["DenseNet264"], use_ssld=use_ssld)
return model

95
ppcls/arch/backbone/model_zoo/distilled_vision_transformer.py
View File

@ -19,15 +19,23 @@ from .vision_transformer import VisionTransformer, Identity, trunc_normal_, zero
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"DeiT_tiny_patch16_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_tiny_patch16_224_pretrained.pdparams",
"DeiT_small_patch16_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_small_patch16_224_pretrained.pdparams",
"DeiT_base_patch16_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_patch16_224_pretrained.pdparams",
"DeiT_tiny_distilled_patch16_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_tiny_distilled_patch16_224_pretrained.pdparams",
"DeiT_small_distilled_patch16_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_small_distilled_patch16_224_pretrained.pdparams",
"DeiT_base_distilled_patch16_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_distilled_patch16_224_pretrained.pdparams",
"DeiT_base_patch16_384": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_patch16_384_pretrained.pdparams",
"DeiT_base_distilled_patch16_384": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_distilled_patch16_384_pretrained.pdparams",
}
"DeiT_tiny_patch16_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_tiny_patch16_224_pretrained.pdparams",
"DeiT_small_patch16_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_small_patch16_224_pretrained.pdparams",
"DeiT_base_patch16_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_patch16_224_pretrained.pdparams",
"DeiT_tiny_distilled_patch16_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_tiny_distilled_patch16_224_pretrained.pdparams",
"DeiT_small_distilled_patch16_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_small_distilled_patch16_224_pretrained.pdparams",
"DeiT_base_distilled_patch16_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_distilled_patch16_224_pretrained.pdparams",
"DeiT_base_patch16_384":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_patch16_384_pretrained.pdparams",
"DeiT_base_distilled_patch16_384":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DeiT_base_distilled_patch16_384_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -36,7 +44,7 @@ class DistilledVisionTransformer(VisionTransformer):
def __init__(self,
img_size=224,
patch_size=16,
class_dim=1000,
class_num=1000,
embed_dim=768,
depth=12,
num_heads=12,
@ -48,7 +56,7 @@ class DistilledVisionTransformer(VisionTransformer):
super().__init__(
img_size=img_size,
patch_size=patch_size,
class_dim=class_dim,
class_num=class_num,
embed_dim=embed_dim,
depth=depth,
num_heads=num_heads,
@ -68,7 +76,7 @@ class DistilledVisionTransformer(VisionTransformer):
self.head_dist = nn.Linear(
self.embed_dim,
self.class_dim) if self.class_dim > 0 else Identity()
self.class_num) if self.class_num > 0 else Identity()
trunc_normal_(self.dist_token)
trunc_normal_(self.pos_embed)
@ -109,7 +117,7 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def DeiT_tiny_patch16_224(pretrained=False, use_ssld=False, **kwargs):
model = VisionTransformer(
@ -121,7 +129,11 @@ def DeiT_tiny_patch16_224(pretrained=False, use_ssld=False, **kwargs):
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_tiny_patch16_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_tiny_patch16_224"],
use_ssld=use_ssld)
return model
@ -135,7 +147,11 @@ def DeiT_small_patch16_224(pretrained=False, use_ssld=False, **kwargs):
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_small_patch16_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_small_patch16_224"],
use_ssld=use_ssld)
return model
@ -149,11 +165,16 @@ def DeiT_base_patch16_224(pretrained=False, use_ssld=False, **kwargs):
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_base_patch16_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_base_patch16_224"],
use_ssld=use_ssld)
return model
def DeiT_tiny_distilled_patch16_224(pretrained=False, use_ssld=False, **kwargs):
def DeiT_tiny_distilled_patch16_224(pretrained=False, use_ssld=False,
**kwargs):
model = DistilledVisionTransformer(
patch_size=16,
embed_dim=192,
@ -163,11 +184,17 @@ def DeiT_tiny_distilled_patch16_224(pretrained=False, use_ssld=False, **kwargs):
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_tiny_distilled_patch16_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_tiny_distilled_patch16_224"],
use_ssld=use_ssld)
return model
def DeiT_small_distilled_patch16_224(pretrained=False, use_ssld=False, **kwargs):
def DeiT_small_distilled_patch16_224(pretrained=False,
use_ssld=False,
**kwargs):
model = DistilledVisionTransformer(
patch_size=16,
embed_dim=384,
@ -177,11 +204,16 @@ def DeiT_small_distilled_patch16_224(pretrained=False, use_ssld=False, **kwargs)
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_small_distilled_patch16_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_small_distilled_patch16_224"],
use_ssld=use_ssld)
return model
def DeiT_base_distilled_patch16_224(pretrained=False, use_ssld=False, **kwargs):
def DeiT_base_distilled_patch16_224(pretrained=False, use_ssld=False,
**kwargs):
model = DistilledVisionTransformer(
patch_size=16,
embed_dim=768,
@ -191,7 +223,11 @@ def DeiT_base_distilled_patch16_224(pretrained=False, use_ssld=False, **kwargs):
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_base_distilled_patch16_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_base_distilled_patch16_224"],
use_ssld=use_ssld)
return model
@ -206,11 +242,16 @@ def DeiT_base_patch16_384(pretrained=False, use_ssld=False, **kwargs):
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_base_patch16_384"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_base_patch16_384"],
use_ssld=use_ssld)
return model
def DeiT_base_distilled_patch16_384(pretrained=False, use_ssld=False, **kwargs):
def DeiT_base_distilled_patch16_384(pretrained=False, use_ssld=False,
**kwargs):
model = DistilledVisionTransformer(
img_size=384,
patch_size=16,
@ -221,5 +262,9 @@ def DeiT_base_distilled_patch16_384(pretrained=False, use_ssld=False, **kwargs):
qkv_bias=True,
epsilon=1e-6,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DeiT_base_distilled_patch16_384"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["DeiT_base_distilled_patch16_384"],
use_ssld=use_ssld)
return model

357
ppcls/arch/backbone/model_zoo/dla.py
View File

@ -23,7 +23,6 @@ from paddle.nn.initializer import Normal, Constant
from ppcls.arch.backbone.base.theseus_layer import Identity
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"DLA34":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DLA34_pretrained.pdparams",
@ -47,10 +46,8 @@ MODEL_URLS = {
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DLA169_pretrained.pdparams"
}
__all__ = MODEL_URLS.keys()
zeros_ = Constant(value=0.)
ones_ = Constant(value=1.)
@ -59,15 +56,23 @@ class DlaBasic(nn.Layer):
def __init__(self, inplanes, planes, stride=1, dilation=1, **cargs):
super(DlaBasic, self).__init__()
self.conv1 = nn.Conv2D(
inplanes, planes, kernel_size=3, stride=stride,
padding=dilation, bias_attr=False, dilation=dilation
)
inplanes,
planes,
kernel_size=3,
stride=stride,
padding=dilation,
bias_attr=False,
dilation=dilation)
self.bn1 = nn.BatchNorm2D(planes)
self.relu = nn.ReLU()
self.conv2 = nn.Conv2D(
planes, planes, kernel_size=3, stride=1,
padding=dilation, bias_attr=False, dilation=dilation
)
planes,
planes,
kernel_size=3,
stride=1,
padding=dilation,
bias_attr=False,
dilation=dilation)
self.bn2 = nn.BatchNorm2D(planes)
self.stride = stride
@ -91,23 +96,34 @@ class DlaBasic(nn.Layer):
class DlaBottleneck(nn.Layer):
expansion = 2
def __init__(self, inplanes, outplanes, stride=1,
dilation=1, cardinality=1, base_width=64):
def __init__(self,
inplanes,
outplanes,
stride=1,
dilation=1,
cardinality=1,
base_width=64):
super(DlaBottleneck, self).__init__()
self.stride = stride
mid_planes = int(math.floor(
outplanes * (base_width / 64)) * cardinality)
mid_planes = int(
math.floor(outplanes * (base_width / 64)) * cardinality)
mid_planes = mid_planes // self.expansion
self.conv1 = nn.Conv2D(inplanes, mid_planes, kernel_size=1, bias_attr=False)
self.conv1 = nn.Conv2D(
inplanes, mid_planes, kernel_size=1, bias_attr=False)
self.bn1 = nn.BatchNorm2D(mid_planes)
self.conv2 = nn.Conv2D(
mid_planes, mid_planes, kernel_size=3,
stride=stride, padding=dilation, bias_attr=False,
dilation=dilation, groups=cardinality
)
mid_planes,
mid_planes,
kernel_size=3,
stride=stride,
padding=dilation,
bias_attr=False,
dilation=dilation,
groups=cardinality)
self.bn2 = nn.BatchNorm2D(mid_planes)
self.conv3 = nn.Conv2D(mid_planes, outplanes, kernel_size=1, bias_attr=False)
self.conv3 = nn.Conv2D(
mid_planes, outplanes, kernel_size=1, bias_attr=False)
self.bn3 = nn.BatchNorm2D(outplanes)
self.relu = nn.ReLU()
@ -136,9 +152,12 @@ class DlaRoot(nn.Layer):
def __init__(self, in_channels, out_channels, kernel_size, residual):
super(DlaRoot, self).__init__()
self.conv = nn.Conv2D(
in_channels, out_channels, 1, stride=1,
bias_attr=False, padding=(kernel_size - 1) // 2
)
in_channels,
out_channels,
1,
stride=1,
bias_attr=False,
padding=(kernel_size - 1) // 2)
self.bn = nn.BatchNorm2D(out_channels)
self.relu = nn.ReLU()
self.residual = residual
@ -155,9 +174,18 @@ class DlaRoot(nn.Layer):
class DlaTree(nn.Layer):
def __init__(self, levels, block, in_channels, out_channels,
stride=1,dilation=1, cardinality=1, base_width=64,
level_root=False, root_dim=0, root_kernel_size=1,
def __init__(self,
levels,
block,
in_channels,
out_channels,
stride=1,
dilation=1,
cardinality=1,
base_width=64,
level_root=False,
root_dim=0,
root_kernel_size=1,
root_residual=False):
super(DlaTree, self).__init__()
if root_dim == 0:
@ -168,28 +196,45 @@ class DlaTree(nn.Layer):
self.downsample = nn.MaxPool2D(
stride, stride=stride) if stride > 1 else Identity()
self.project = Identity()
cargs = dict(dilation=dilation, cardinality=cardinality, base_width=base_width)
cargs = dict(
dilation=dilation, cardinality=cardinality, base_width=base_width)
if levels == 1:
self.tree1 = block(in_channels, out_channels, stride, **cargs)
self.tree2 = block(out_channels, out_channels, 1, **cargs)
if in_channels != out_channels:
self.project = nn.Sequential(
nn.Conv2D(in_channels, out_channels, kernel_size=1, stride=1, bias_attr=False),
nn.Conv2D(
in_channels,
out_channels,
kernel_size=1,
stride=1,
bias_attr=False),
nn.BatchNorm2D(out_channels))
else:
cargs.update(dict(root_kernel_size=root_kernel_size, root_residual=root_residual))
cargs.update(
dict(
root_kernel_size=root_kernel_size,
root_residual=root_residual))
self.tree1 = DlaTree(
levels - 1, block, in_channels,
out_channels, stride, root_dim=0, **cargs
)
levels - 1,
block,
in_channels,
out_channels,
stride,
root_dim=0,
**cargs)
self.tree2 = DlaTree(
levels - 1, block, out_channels,
out_channels, root_dim=root_dim + out_channels, **cargs
)
levels - 1,
block,
out_channels,
out_channels,
root_dim=root_dim + out_channels,
**cargs)
if levels == 1:
self.root = DlaRoot(root_dim, out_channels, root_kernel_size, root_residual)
self.root = DlaRoot(root_dim, out_channels, root_kernel_size,
root_residual)
self.level_root = level_root
self.root_dim = root_dim
@ -214,12 +259,20 @@ class DlaTree(nn.Layer):
class DLA(nn.Layer):
def __init__(self, levels, channels, in_chans=3, cardinality=1,
base_width=64, block=DlaBottleneck, residual_root=False,
drop_rate=0.0, class_dim=1000, with_pool=True):
def __init__(self,
levels,
channels,
in_chans=3,
cardinality=1,
base_width=64,
block=DlaBottleneck,
residual_root=False,
drop_rate=0.0,
class_num=1000,
with_pool=True):
super(DLA, self).__init__()
self.channels = channels
self.class_dim = class_dim
self.class_num = class_num
self.with_pool = with_pool
self.cardinality = cardinality
self.base_width = base_width
@ -227,46 +280,72 @@ class DLA(nn.Layer):
self.base_layer = nn.Sequential(
nn.Conv2D(
in_chans, channels[0], kernel_size=7,
stride=1, padding=3, bias_attr=False
),
in_chans,
channels[0],
kernel_size=7,
stride=1,
padding=3,
bias_attr=False),
nn.BatchNorm2D(channels[0]),
nn.ReLU())
self.level0 = self._make_conv_level(channels[0], channels[0], levels[0])
self.level1 = self._make_conv_level(channels[0], channels[1], levels[1], stride=2)
self.level0 = self._make_conv_level(channels[0], channels[0],
levels[0])
self.level1 = self._make_conv_level(
channels[0], channels[1], levels[1], stride=2)
cargs = dict(
cardinality=cardinality,
base_width=base_width,
root_residual=residual_root
)
cardinality=cardinality,
base_width=base_width,
root_residual=residual_root)
self.level2 = DlaTree(
levels[2], block, channels[1],
channels[2], 2, level_root=False, **cargs
)
levels[2],
block,
channels[1],
channels[2],
2,
level_root=False,
**cargs)
self.level3 = DlaTree(
levels[3], block, channels[2],
channels[3], 2, level_root=True, **cargs
)
levels[3],
block,
channels[2],
channels[3],
2,
level_root=True,
**cargs)
self.level4 = DlaTree(
levels[4], block, channels[3],
channels[4], 2, level_root=True, **cargs
)
levels[4],
block,
channels[3],
channels[4],
2,
level_root=True,
**cargs)
self.level5 = DlaTree(
levels[5], block, channels[4],
channels[5], 2, level_root=True, **cargs
)
levels[5],
block,
channels[4],
channels[5],
2,
level_root=True,
**cargs)
self.feature_info = [
# rare to have a meaningful stride 1 level
dict(num_chs=channels[0], reduction=1, module='level0'),
dict(num_chs=channels[1], reduction=2, module='level1'),
dict(num_chs=channels[2], reduction=4, module='level2'),
dict(num_chs=channels[3], reduction=8, module='level3'),
dict(num_chs=channels[4], reduction=16, module='level4'),
dict(num_chs=channels[5], reduction=32, module='level5'),
dict(
num_chs=channels[0], reduction=1, module='level0'),
dict(
num_chs=channels[1], reduction=2, module='level1'),
dict(
num_chs=channels[2], reduction=4, module='level2'),
dict(
num_chs=channels[3], reduction=8, module='level3'),
dict(
num_chs=channels[4], reduction=16, module='level4'),
dict(
num_chs=channels[5], reduction=32, module='level5'),
]
self.num_features = channels[-1]
@ -274,8 +353,8 @@ class DLA(nn.Layer):
if with_pool:
self.global_pool = nn.AdaptiveAvgPool2D(1)
if class_dim > 0:
self.fc = nn.Conv2D(self.num_features, class_dim, 1)
if class_num > 0:
self.fc = nn.Conv2D(self.num_features, class_num, 1)
for m in self.sublayers():
if isinstance(m, nn.Conv2D):
@ -291,12 +370,14 @@ class DLA(nn.Layer):
for i in range(convs):
modules.extend([
nn.Conv2D(
inplanes, planes, kernel_size=3,
inplanes,
planes,
kernel_size=3,
stride=stride if i == 0 else 1,
padding=dilation, bias_attr=False, dilation=dilation
),
nn.BatchNorm2D(planes),
nn.ReLU()])
padding=dilation,
bias_attr=False,
dilation=dilation), nn.BatchNorm2D(planes), nn.ReLU()
])
inplanes = planes
return nn.Sequential(*modules)
@ -321,7 +402,7 @@ class DLA(nn.Layer):
if self.drop_rate > 0.:
x = F.dropout(x, p=self.drop_rate, training=self.training)
if self.class_dim > 0:
if self.class_num > 0:
x = self.fc(x)
x = x.flatten(1)
@ -342,124 +423,104 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def DLA34(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 2, 2, 1),
channels=(16, 32, 64, 128, 256, 512),
block=DlaBasic,
**kwargs
)
model = DLA(levels=(1, 1, 1, 2, 2, 1),
channels=(16, 32, 64, 128, 256, 512),
block=DlaBasic,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA34"])
return model
def DLA46_c(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 2, 2, 1),
channels=(16, 32, 64, 64, 128, 256),
block=DlaBottleneck,
**kwargs
)
model = DLA(levels=(1, 1, 1, 2, 2, 1),
channels=(16, 32, 64, 64, 128, 256),
block=DlaBottleneck,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA46_c"])
return model
def DLA46x_c(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 2, 2, 1),
channels=(16, 32, 64, 64, 128, 256),
block=DlaBottleneck,
cardinality=32,
base_width=4,
**kwargs
)
model = DLA(levels=(1, 1, 1, 2, 2, 1),
channels=(16, 32, 64, 64, 128, 256),
block=DlaBottleneck,
cardinality=32,
base_width=4,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA46x_c"])
return model
def DLA60(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 2, 3, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
**kwargs
)
model = DLA(levels=(1, 1, 1, 2, 3, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA60"])
return model
def DLA60x(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 2, 3, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
cardinality=32,
base_width=4,
**kwargs
)
model = DLA(levels=(1, 1, 1, 2, 3, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
cardinality=32,
base_width=4,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA60x"])
return model
def DLA60x_c(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 2, 3, 1),
channels=(16, 32, 64, 64, 128, 256),
block=DlaBottleneck,
cardinality=32,
base_width=4,
**kwargs
)
model = DLA(levels=(1, 1, 1, 2, 3, 1),
channels=(16, 32, 64, 64, 128, 256),
block=DlaBottleneck,
cardinality=32,
base_width=4,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA60x_c"])
return model
def DLA102(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 3, 4, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
residual_root=True,
**kwargs
)
model = DLA(levels=(1, 1, 1, 3, 4, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
residual_root=True,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA102"])
return model
def DLA102x(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 3, 4, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
cardinality=32,
base_width=4,
residual_root=True,
**kwargs
)
model = DLA(levels=(1, 1, 1, 3, 4, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
cardinality=32,
base_width=4,
residual_root=True,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA102x"])
return model
def DLA102x2(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 1, 3, 4, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
cardinality=64,
base_width=4,
residual_root=True,
**kwargs
)
model = DLA(levels=(1, 1, 1, 3, 4, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
cardinality=64,
base_width=4,
residual_root=True,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA102x2"])
return model
def DLA169(pretrained=False, **kwargs):
model = DLA(
levels=(1, 1, 2, 3, 5, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
residual_root=True,
**kwargs
)
model = DLA(levels=(1, 1, 2, 3, 5, 1),
channels=(16, 32, 128, 256, 512, 1024),
block=DlaBottleneck,
residual_root=True,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DLA169"])
return model

31
ppcls/arch/backbone/model_zoo/dpn.py
View File

@ -29,12 +29,18 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"DPN68": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN68_pretrained.pdparams",
"DPN92": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN92_pretrained.pdparams",
"DPN98": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN98_pretrained.pdparams",
"DPN107": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN107_pretrained.pdparams",
"DPN131": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN131_pretrained.pdparams",
}
MODEL_URLS = {
"DPN68":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN68_pretrained.pdparams",
"DPN92":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN92_pretrained.pdparams",
"DPN98":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN98_pretrained.pdparams",
"DPN107":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN107_pretrained.pdparams",
"DPN131":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/DPN131_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -211,10 +217,10 @@ class DualPathFactory(nn.Layer):
class DPN(nn.Layer):
def __init__(self, layers=68, class_dim=1000):
def __init__(self, layers=68, class_num=1000):
super(DPN, self).__init__()
self._class_dim = class_dim
self._class_num = class_num
args = self.get_net_args(layers)
bws = args['bw']
@ -309,7 +315,7 @@ class DPN(nn.Layer):
self.out = Linear(
out_channel,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_weights"),
bias_attr=ParamAttr(name="fc_offset"))
@ -400,7 +406,8 @@ class DPN(nn.Layer):
net_arg['init_padding'] = init_padding
return net_arg
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -411,7 +418,7 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
else:
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
)
def DPN68(pretrained=False, use_ssld=False, **kwargs):
@ -441,4 +448,4 @@ def DPN107(pretrained=False, use_ssld=False, **kwargs):
def DPN131(pretrained=False, use_ssld=False, **kwargs):
model = DPN(layers=131, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["DPN131"])
return model
return model

76
ppcls/arch/backbone/model_zoo/efficientnet.py
View File

@ -11,16 +11,26 @@ import copy
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"EfficientNetB0_small": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB0_small_pretrained.pdparams",
"EfficientNetB0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB0_pretrained.pdparams",
"EfficientNetB1": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB1_pretrained.pdparams",
"EfficientNetB2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB2_pretrained.pdparams",
"EfficientNetB3": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB3_pretrained.pdparams",
"EfficientNetB4": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB4_pretrained.pdparams",
"EfficientNetB5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB5_pretrained.pdparams",
"EfficientNetB6": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB6_pretrained.pdparams",
"EfficientNetB7": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB7_pretrained.pdparams",
}
MODEL_URLS = {
"EfficientNetB0_small":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB0_small_pretrained.pdparams",
"EfficientNetB0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB0_pretrained.pdparams",
"EfficientNetB1":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB1_pretrained.pdparams",
"EfficientNetB2":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB2_pretrained.pdparams",
"EfficientNetB3":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB3_pretrained.pdparams",
"EfficientNetB4":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB4_pretrained.pdparams",
"EfficientNetB5":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB5_pretrained.pdparams",
"EfficientNetB6":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB6_pretrained.pdparams",
"EfficientNetB7":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/EfficientNetB7_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -725,7 +735,7 @@ class EfficientNet(nn.Layer):
padding_type="SAME",
override_params=None,
use_se=True,
class_dim=1000):
class_num=1000):
super(EfficientNet, self).__init__()
model_name = 'efficientnet-' + name
@ -778,7 +788,7 @@ class EfficientNet(nn.Layer):
param_attr, bias_attr = init_fc_layer("_fc")
self._fc = Linear(
output_channels,
class_dim,
class_num,
weight_attr=param_attr,
bias_attr=bias_attr)
@ -792,7 +802,7 @@ class EfficientNet(nn.Layer):
x = self._fc(x)
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -803,14 +813,14 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
else:
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
)
def EfficientNetB0_small(padding_type='DYNAMIC',
override_params=None,
use_se=False,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b0',
@ -825,8 +835,8 @@ def EfficientNetB0_small(padding_type='DYNAMIC',
def EfficientNetB0(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b0',
@ -841,8 +851,8 @@ def EfficientNetB0(padding_type='SAME',
def EfficientNetB1(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b1',
@ -857,8 +867,8 @@ def EfficientNetB1(padding_type='SAME',
def EfficientNetB2(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b2',
@ -873,8 +883,8 @@ def EfficientNetB2(padding_type='SAME',
def EfficientNetB3(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b3',
@ -889,8 +899,8 @@ def EfficientNetB3(padding_type='SAME',
def EfficientNetB4(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b4',
@ -905,8 +915,8 @@ def EfficientNetB4(padding_type='SAME',
def EfficientNetB5(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b5',
@ -921,8 +931,8 @@ def EfficientNetB5(padding_type='SAME',
def EfficientNetB6(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b6',
@ -937,8 +947,8 @@ def EfficientNetB6(padding_type='SAME',
def EfficientNetB7(padding_type='SAME',
override_params=None,
use_se=True,
pretrained=False,
use_ssld=False,
pretrained=False,
use_ssld=False,
**kwargs):
model = EfficientNet(
name='b7',
@ -947,4 +957,4 @@ def EfficientNetB7(padding_type='SAME',
use_se=use_se,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["EfficientNetB7"])
return model
return model

27
ppcls/arch/backbone/model_zoo/ghostnet.py
View File

@ -23,10 +23,14 @@ from paddle.nn.initializer import Uniform, KaimingNormal
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"GhostNet_x0_5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GhostNet_x0_5_pretrained.pdparams",
"GhostNet_x1_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GhostNet_x1_0_pretrained.pdparams",
"GhostNet_x1_3": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GhostNet_x1_3_pretrained.pdparams",
}
MODEL_URLS = {
"GhostNet_x0_5":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GhostNet_x0_5_pretrained.pdparams",
"GhostNet_x1_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GhostNet_x1_0_pretrained.pdparams",
"GhostNet_x1_3":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GhostNet_x1_3_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -215,7 +219,7 @@ class GhostBottleneck(nn.Layer):
class GhostNet(nn.Layer):
def __init__(self, scale, class_dim=1000):
def __init__(self, scale, class_num=1000):
super(GhostNet, self).__init__()
self.cfgs = [
# k, t, c, SE, s
@ -290,7 +294,7 @@ class GhostNet(nn.Layer):
stdv = 1.0 / math.sqrt(self._fc0_output_channels * 1.0)
self.fc_1 = Linear(
self._fc0_output_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
name="fc_1_weights", initializer=Uniform(-stdv, stdv)),
bias_attr=ParamAttr(name="fc_1_offset"))
@ -322,7 +326,7 @@ class GhostNet(nn.Layer):
new_v += divisor
return new_v
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -338,17 +342,20 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def GhostNet_x0_5(pretrained=False, use_ssld=False, **kwargs):
model = GhostNet(scale=0.5, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["GhostNet_x0_5"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["GhostNet_x0_5"], use_ssld=use_ssld)
return model
def GhostNet_x1_0(pretrained=False, use_ssld=False, **kwargs):
model = GhostNet(scale=1.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["GhostNet_x1_0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["GhostNet_x1_0"], use_ssld=use_ssld)
return model
def GhostNet_x1_3(pretrained=False, use_ssld=False, **kwargs):
model = GhostNet(scale=1.3, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["GhostNet_x1_3"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["GhostNet_x1_3"], use_ssld=use_ssld)
return model

21
ppcls/arch/backbone/model_zoo/googlenet.py
View File

@ -10,8 +10,10 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"GoogLeNet": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GoogLeNet_pretrained.pdparams",
}
MODEL_URLS = {
"GoogLeNet":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/GoogLeNet_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -101,7 +103,7 @@ class Inception(nn.Layer):
class GoogLeNetDY(nn.Layer):
def __init__(self, class_dim=1000):
def __init__(self, class_num=1000):
super(GoogLeNetDY, self).__init__()
self._conv = ConvLayer(3, 64, 7, 2, name="conv1")
self._pool = MaxPool2D(kernel_size=3, stride=2)
@ -134,7 +136,7 @@ class GoogLeNetDY(nn.Layer):
self._drop = Dropout(p=0.4, mode="downscale_in_infer")
self._fc_out = Linear(
1024,
class_dim,
class_num,
weight_attr=xavier(1024, 1, "out"),
bias_attr=ParamAttr(name="out_offset"))
self._pool_o1 = AvgPool2D(kernel_size=5, stride=3)
@ -147,7 +149,7 @@ class GoogLeNetDY(nn.Layer):
self._drop_o1 = Dropout(p=0.7, mode="downscale_in_infer")
self._out1 = Linear(
1024,
class_dim,
class_num,
weight_attr=xavier(1024, 1, "out1"),
bias_attr=ParamAttr(name="out1_offset"))
self._pool_o2 = AvgPool2D(kernel_size=5, stride=3)
@ -160,7 +162,7 @@ class GoogLeNetDY(nn.Layer):
self._drop_o2 = Dropout(p=0.7, mode="downscale_in_infer")
self._out2 = Linear(
1024,
class_dim,
class_num,
weight_attr=xavier(1024, 1, "out2"),
bias_attr=ParamAttr(name="out2_offset"))
@ -205,8 +207,8 @@ class GoogLeNetDY(nn.Layer):
x = self._drop_o2(x)
out2 = self._out2(x)
return [out, out1, out2]
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -222,5 +224,6 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def GoogLeNet(pretrained=False, use_ssld=False, **kwargs):
model = GoogLeNetDY(**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["GoogLeNet"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["GoogLeNet"], use_ssld=use_ssld)
return model

44
ppcls/arch/backbone/model_zoo/gvt.py
View File

@ -25,18 +25,23 @@ from .vision_transformer import Block as ViTBlock
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"pcpvt_small": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/pcpvt_small_pretrained.pdparams",
"pcpvt_base": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/pcpvt_base_pretrained.pdparams",
"pcpvt_large": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/pcpvt_large_pretrained.pdparams",
"alt_gvt_small": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/alt_gvt_small_pretrained.pdparams",
"alt_gvt_base": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/alt_gvt_base_pretrained.pdparams",
"alt_gvt_large": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/alt_gvt_large_pretrained.pdparams"
}
"pcpvt_small":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/pcpvt_small_pretrained.pdparams",
"pcpvt_base":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/pcpvt_base_pretrained.pdparams",
"pcpvt_large":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/pcpvt_large_pretrained.pdparams",
"alt_gvt_small":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/alt_gvt_small_pretrained.pdparams",
"alt_gvt_base":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/alt_gvt_base_pretrained.pdparams",
"alt_gvt_large":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/alt_gvt_large_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
class GroupAttention(nn.Layer):
"""LSA: self attention within a group.
"""
@ -522,7 +527,7 @@ class ALTGVT(PCPVT):
img_size=224,
patch_size=4,
in_chans=3,
class_dim=1000,
class_num=1000,
embed_dims=[64, 128, 256],
num_heads=[1, 2, 4],
mlp_ratios=[4, 4, 4],
@ -536,7 +541,7 @@ class ALTGVT(PCPVT):
sr_ratios=[4, 2, 1],
block_cls=GroupBlock,
wss=[7, 7, 7]):
super().__init__(img_size, patch_size, in_chans, class_dim, embed_dims,
super().__init__(img_size, patch_size, in_chans, class_num, embed_dims,
num_heads, mlp_ratios, qkv_bias, qk_scale, drop_rate,
attn_drop_rate, drop_path_rate, norm_layer, depths,
sr_ratios, block_cls)
@ -568,6 +573,7 @@ class ALTGVT(PCPVT):
cur += depths[k]
self.apply(self._init_weights)
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -593,7 +599,8 @@ def pcpvt_small(pretrained=False, use_ssld=False, **kwargs):
depths=[3, 4, 6, 3],
sr_ratios=[8, 4, 2, 1],
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["pcpvt_small"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["pcpvt_small"], use_ssld=use_ssld)
return model
@ -609,7 +616,8 @@ def pcpvt_base(pretrained=False, use_ssld=False, **kwargs):
depths=[3, 4, 18, 3],
sr_ratios=[8, 4, 2, 1],
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["pcpvt_base"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["pcpvt_base"], use_ssld=use_ssld)
return model
@ -625,7 +633,8 @@ def pcpvt_large(pretrained=False, use_ssld=False, **kwargs):
depths=[3, 8, 27, 3],
sr_ratios=[8, 4, 2, 1],
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["pcpvt_large"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["pcpvt_large"], use_ssld=use_ssld)
return model
@ -642,7 +651,8 @@ def alt_gvt_small(pretrained=False, use_ssld=False, **kwargs):
wss=[7, 7, 7, 7],
sr_ratios=[8, 4, 2, 1],
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["alt_gvt_small"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["alt_gvt_small"], use_ssld=use_ssld)
return model
@ -659,7 +669,8 @@ def alt_gvt_base(pretrained=False, use_ssld=False, **kwargs):
wss=[7, 7, 7, 7],
sr_ratios=[8, 4, 2, 1],
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["alt_gvt_base"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["alt_gvt_base"], use_ssld=use_ssld)
return model
@ -676,5 +687,6 @@ def alt_gvt_large(pretrained=False, use_ssld=False, **kwargs):
wss=[7, 7, 7, 7],
sr_ratios=[8, 4, 2, 1],
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["alt_gvt_large"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["alt_gvt_large"], use_ssld=use_ssld)
return model

106
ppcls/arch/backbone/model_zoo/hardnet.py
View File

@ -17,7 +17,6 @@ import paddle.nn as nn
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
'HarDNet39_ds':
'https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/HarDNet39_ds_pretrained.pdparams',
@ -29,51 +28,70 @@ MODEL_URLS = {
'https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/HarDNet85_pretrained.pdparams'
}
__all__ = MODEL_URLS.keys()
def ConvLayer(in_channels, out_channels, kernel_size=3, stride=1, bias_attr=False):
def ConvLayer(in_channels,
out_channels,
kernel_size=3,
stride=1,
bias_attr=False):
layer = nn.Sequential(
('conv', nn.Conv2D(
in_channels, out_channels, kernel_size=kernel_size,
stride=stride, padding=kernel_size//2, groups=1, bias_attr=bias_attr
)),
('norm', nn.BatchNorm2D(out_channels)),
('relu', nn.ReLU6())
)
in_channels,
out_channels,
kernel_size=kernel_size,
stride=stride,
padding=kernel_size // 2,
groups=1,
bias_attr=bias_attr)), ('norm', nn.BatchNorm2D(out_channels)),
('relu', nn.ReLU6()))
return layer
def DWConvLayer(in_channels, out_channels, kernel_size=3, stride=1, bias_attr=False):
def DWConvLayer(in_channels,
out_channels,
kernel_size=3,
stride=1,
bias_attr=False):
layer = nn.Sequential(
('dwconv', nn.Conv2D(
in_channels, out_channels, kernel_size=kernel_size,
stride=stride, padding=1, groups=out_channels, bias_attr=bias_attr
)),
('norm', nn.BatchNorm2D(out_channels))
)
in_channels,
out_channels,
kernel_size=kernel_size,
stride=stride,
padding=1,
groups=out_channels,
bias_attr=bias_attr)), ('norm', nn.BatchNorm2D(out_channels)))
return layer
def CombConvLayer(in_channels, out_channels, kernel_size=1, stride=1):
layer = nn.Sequential(
('layer1', ConvLayer(in_channels, out_channels, kernel_size=kernel_size)),
('layer2', DWConvLayer(out_channels, out_channels, stride=stride))
)
('layer1', ConvLayer(
in_channels, out_channels, kernel_size=kernel_size)),
('layer2', DWConvLayer(
out_channels, out_channels, stride=stride)))
return layer
class HarDBlock(nn.Layer):
def __init__(self, in_channels, growth_rate, grmul, n_layers,
keepBase=False, residual_out=False, dwconv=False):
def __init__(self,
in_channels,
growth_rate,
grmul,
n_layers,
keepBase=False,
residual_out=False,
dwconv=False):
super().__init__()
self.keepBase = keepBase
self.links = []
layers_ = []
self.out_channels = 0 # if upsample else in_channels
for i in range(n_layers):
outch, inch, link = self.get_link(i+1, in_channels, growth_rate, grmul)
outch, inch, link = self.get_link(i + 1, in_channels, growth_rate,
grmul)
self.links.append(link)
if dwconv:
layers_.append(CombConvLayer(inch, outch))
@ -92,7 +110,7 @@ class HarDBlock(nn.Layer):
link = []
for i in range(10):
dv = 2 ** i
dv = 2**i
if layer % dv == 0:
k = layer - dv
link.append(k)
@ -126,7 +144,7 @@ class HarDBlock(nn.Layer):
t = len(layers_)
out_ = []
for i in range(t):
if (i == 0 and self.keepBase) or (i == t-1) or (i % 2 == 1):
if (i == 0 and self.keepBase) or (i == t - 1) or (i % 2 == 1):
out_.append(layers_[i])
out = paddle.concat(out_, 1)
@ -134,8 +152,11 @@ class HarDBlock(nn.Layer):
class HarDNet(nn.Layer):
def __init__(self, depth_wise=False, arch=85,
class_dim=1000, with_pool=True):
def __init__(self,
depth_wise=False,
arch=85,
class_num=1000,
with_pool=True):
super().__init__()
first_ch = [32, 64]
second_kernel = 3
@ -146,16 +167,16 @@ class HarDNet(nn.Layer):
# HarDNet68
ch_list = [128, 256, 320, 640, 1024]
gr = [14, 16, 20, 40, 160]
n_layers = [8, 16, 16, 16, 4]
downSamp = [1, 0, 1, 1, 0]
n_layers = [8, 16, 16, 16, 4]
downSamp = [1, 0, 1, 1, 0]
if arch == 85:
# HarDNet85
first_ch = [48, 96]
ch_list = [192, 256, 320, 480, 720, 1280]
gr = [24, 24, 28, 36, 48, 256]
n_layers = [8, 16, 16, 16, 16, 4]
downSamp = [1, 0, 1, 0, 1, 0]
gr = [24, 24, 28, 36, 48, 256]
n_layers = [8, 16, 16, 16, 16, 4]
downSamp = [1, 0, 1, 0, 1, 0]
drop_rate = 0.2
elif arch == 39:
@ -163,9 +184,9 @@ class HarDNet(nn.Layer):
first_ch = [24, 48]
ch_list = [96, 320, 640, 1024]
grmul = 1.6
gr = [16, 20, 64, 160]
n_layers = [4, 16, 8, 4]
downSamp = [1, 1, 1, 0]
gr = [16, 20, 64, 160]
n_layers = [4, 16, 8, 4]
downSamp = [1, 1, 1, 0]
if depth_wise:
second_kernel = 1
@ -177,12 +198,17 @@ class HarDNet(nn.Layer):
# First Layer: Standard Conv3x3, Stride=2
self.base.append(
ConvLayer(in_channels=3, out_channels=first_ch[0], kernel_size=3,
stride=2, bias_attr=False))
ConvLayer(
in_channels=3,
out_channels=first_ch[0],
kernel_size=3,
stride=2,
bias_attr=False))
# Second Layer
self.base.append(
ConvLayer(first_ch[0], first_ch[1], kernel_size=second_kernel))
ConvLayer(
first_ch[0], first_ch[1], kernel_size=second_kernel))
# Maxpooling or DWConv3x3 downsampling
if max_pool:
@ -197,7 +223,7 @@ class HarDNet(nn.Layer):
ch = blk.out_channels
self.base.append(blk)
if i == blks-1 and arch == 85:
if i == blks - 1 and arch == 85:
self.base.append(nn.Dropout(0.1))
self.base.append(ConvLayer(ch, ch_list[i], kernel_size=1))
@ -208,17 +234,17 @@ class HarDNet(nn.Layer):
else:
self.base.append(DWConvLayer(ch, ch, stride=2))
ch = ch_list[blks-1]
ch = ch_list[blks - 1]
layers = []
if with_pool:
layers.append(nn.AdaptiveAvgPool2D((1, 1)))
if class_dim > 0:
if class_num > 0:
layers.append(nn.Flatten())
layers.append(nn.Dropout(drop_rate))
layers.append(nn.Linear(ch, class_dim))
layers.append(nn.Linear(ch, class_num))
self.base.append(nn.Sequential(*layers))

13
ppcls/arch/backbone/model_zoo/inception_v4.py
View File

@ -23,7 +23,10 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"InceptionV4": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/InceptionV4_pretrained.pdparams"}
MODEL_URLS = {
"InceptionV4":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/InceptionV4_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
@ -392,7 +395,7 @@ class InceptionC(nn.Layer):
class InceptionV4DY(nn.Layer):
def __init__(self, class_dim=1000):
def __init__(self, class_num=1000):
super(InceptionV4DY, self).__init__()
self._inception_stem = InceptionStem()
@ -420,7 +423,7 @@ class InceptionV4DY(nn.Layer):
stdv = 1.0 / math.sqrt(1536 * 1.0)
self.out = Linear(
1536,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="final_fc_weights"),
bias_attr=ParamAttr(name="final_fc_offset"))
@ -466,7 +469,9 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
"pretrained type is not available. Please use `string` or `boolean` type."
)
def InceptionV4(pretrained=False, use_ssld=False, **kwargs):
model = InceptionV4DY(**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["InceptionV4"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["InceptionV4"], use_ssld=use_ssld)
return model

85
ppcls/arch/backbone/model_zoo/levit.py
View File

@ -27,12 +27,17 @@ from .vision_transformer import trunc_normal_, zeros_, ones_, Identity
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"LeViT_128S": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_128S_pretrained.pdparams",
"LeViT_128": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_128_pretrained.pdparams",
"LeViT_192": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_192_pretrained.pdparams",
"LeViT_256": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_256_pretrained.pdparams",
"LeViT_384": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_384_pretrained.pdparams",
}
"LeViT_128S":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_128S_pretrained.pdparams",
"LeViT_128":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_128_pretrained.pdparams",
"LeViT_192":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_192_pretrained.pdparams",
"LeViT_256":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_256_pretrained.pdparams",
"LeViT_384":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/LeViT_384_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -326,7 +331,7 @@ class LeViT(nn.Layer):
img_size=224,
patch_size=16,
in_chans=3,
class_dim=1000,
class_num=1000,
embed_dim=[192],
key_dim=[64],
depth=[12],
@ -341,7 +346,7 @@ class LeViT(nn.Layer):
drop_path=0):
super().__init__()
self.class_dim = class_dim
self.class_num = class_num
self.num_features = embed_dim[-1]
self.embed_dim = embed_dim
self.distillation = distillation
@ -403,10 +408,10 @@ class LeViT(nn.Layer):
# Classifier head
self.head = BN_Linear(embed_dim[-1],
class_dim) if class_dim > 0 else Identity()
class_num) if class_num > 0 else Identity()
if distillation:
self.head_dist = BN_Linear(
embed_dim[-1], class_dim) if class_dim > 0 else Identity()
embed_dim[-1], class_num) if class_num > 0 else Identity()
def forward(self, x):
x = self.patch_embed(x)
@ -423,7 +428,7 @@ class LeViT(nn.Layer):
return x
def model_factory(C, D, X, N, drop_path, class_dim, distillation):
def model_factory(C, D, X, N, drop_path, class_num, distillation):
embed_dim = [int(x) for x in C.split('_')]
num_heads = [int(x) for x in N.split('_')]
depth = [int(x) for x in X.split('_')]
@ -444,7 +449,7 @@ def model_factory(C, D, X, N, drop_path, class_dim, distillation):
attention_activation=act,
mlp_activation=act,
hybrid_backbone=b16(embed_dim[0], activation=act),
class_dim=class_dim,
class_num=class_num,
drop_path=drop_path,
distillation=distillation)
@ -489,6 +494,7 @@ specification = {
},
}
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -502,46 +508,71 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
)
def LeViT_128S(pretrained=False, use_ssld=False, class_dim=1000, distillation=False, **kwargs):
def LeViT_128S(pretrained=False,
use_ssld=False,
class_num=1000,
distillation=False,
**kwargs):
model = model_factory(
**specification['LeViT_128S'],
class_dim=class_dim,
class_num=class_num,
distillation=distillation)
_load_pretrained(pretrained, model, MODEL_URLS["LeViT_128S"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["LeViT_128S"], use_ssld=use_ssld)
return model
def LeViT_128(pretrained=False, use_ssld=False, class_dim=1000, distillation=False, **kwargs):
def LeViT_128(pretrained=False,
use_ssld=False,
class_num=1000,
distillation=False,
**kwargs):
model = model_factory(
**specification['LeViT_128'],
class_dim=class_dim,
class_num=class_num,
distillation=distillation)
_load_pretrained(pretrained, model, MODEL_URLS["LeViT_128"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["LeViT_128"], use_ssld=use_ssld)
return model
def LeViT_192(pretrained=False, use_ssld=False, class_dim=1000, distillation=False, **kwargs):
def LeViT_192(pretrained=False,
use_ssld=False,
class_num=1000,
distillation=False,
**kwargs):
model = model_factory(
**specification['LeViT_192'],
class_dim=class_dim,
class_num=class_num,
distillation=distillation)
_load_pretrained(pretrained, model, MODEL_URLS["LeViT_192"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["LeViT_192"], use_ssld=use_ssld)
return model
def LeViT_256(pretrained=False, use_ssld=False, class_dim=1000, distillation=False, **kwargs):
def LeViT_256(pretrained=False,
use_ssld=False,
class_num=1000,
distillation=False,
**kwargs):
model = model_factory(
**specification['LeViT_256'],
class_dim=class_dim,
class_num=class_num,
distillation=distillation)
_load_pretrained(pretrained, model, MODEL_URLS["LeViT_256"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["LeViT_256"], use_ssld=use_ssld)
return model
def LeViT_384(pretrained=False, use_ssld=False, class_dim=1000, distillation=False, **kwargs):
def LeViT_384(pretrained=False,
use_ssld=False,
class_num=1000,
distillation=False,
**kwargs):
model = model_factory(
**specification['LeViT_384'],
class_dim=class_dim,
class_num=class_num,
distillation=distillation)
_load_pretrained(pretrained, model, MODEL_URLS["LeViT_384"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["LeViT_384"], use_ssld=use_ssld)
return model

33
ppcls/arch/backbone/model_zoo/mixnet.py
View File

@ -25,9 +25,14 @@ import paddle.nn as nn
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"MixNet_S": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MixNet_S_pretrained.pdparams",
"MixNet_M": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MixNet_M_pretrained.pdparams",
"MixNet_L": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MixNet_L_pretrained.pdparams"}
MODEL_URLS = {
"MixNet_S":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MixNet_S_pretrained.pdparams",
"MixNet_M":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MixNet_M_pretrained.pdparams",
"MixNet_L":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MixNet_L_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
@ -617,7 +622,7 @@ class MixNet(nn.Layer):
Number of input channels.
in_size : tuple of two ints, default (224, 224)
Spatial size of the expected input image.
class_dim : int, default 1000
class_num : int, default 1000
Number of classification classes.
"""
@ -632,10 +637,10 @@ class MixNet(nn.Layer):
se_factors,
in_channels=3,
in_size=(224, 224),
class_dim=1000):
class_num=1000):
super(MixNet, self).__init__()
self.in_size = in_size
self.class_dim = class_dim
self.class_num = class_num
self.features = nn.Sequential()
self.features.add_sublayer(
@ -687,7 +692,7 @@ class MixNet(nn.Layer):
kernel_size=7, stride=1))
self.output = nn.Linear(
in_features=in_channels, out_features=class_dim)
in_features=in_channels, out_features=class_num)
def forward(self, x):
x = self.features(x)
@ -773,9 +778,11 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
"pretrained type is not available. Please use `string` or `boolean` type."
)
def MixNet_S(pretrained=False, use_ssld=False, **kwargs):
model = InceptionV4DY(**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["InceptionV4"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["InceptionV4"], use_ssld=use_ssld)
return model
@ -786,7 +793,8 @@ def MixNet_S(**kwargs):
"""
model = get_mixnet(
version="s", width_scale=1.0, model_name="MixNet_S", **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MixNet_S"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MixNet_S"], use_ssld=use_ssld)
return model
@ -797,7 +805,8 @@ def MixNet_M(**kwargs):
"""
model = get_mixnet(
version="m", width_scale=1.0, model_name="MixNet_M", **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MixNet_M"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MixNet_M"], use_ssld=use_ssld)
return model
@ -808,6 +817,6 @@ def MixNet_L(**kwargs):
"""
model = get_mixnet(
version="m", width_scale=1.3, model_name="MixNet_L", **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MixNet_L"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MixNet_L"], use_ssld=use_ssld)
return model

50
ppcls/arch/backbone/model_zoo/mobilenet_v2.py
View File

@ -28,12 +28,20 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"MobileNetV2_x0_25": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x0_25_pretrained.pdparams",
"MobileNetV2_x0_5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x0_5_pretrained.pdparams",
"MobileNetV2_x0_75": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x0_75_pretrained.pdparams",
"MobileNetV2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_pretrained.pdparams",
"MobileNetV2_x1_5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x1_5_pretrained.pdparams",
"MobileNetV2_x2_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x2_0_pretrained.pdparams"}
MODEL_URLS = {
"MobileNetV2_x0_25":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x0_25_pretrained.pdparams",
"MobileNetV2_x0_5":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x0_5_pretrained.pdparams",
"MobileNetV2_x0_75":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x0_75_pretrained.pdparams",
"MobileNetV2":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_pretrained.pdparams",
"MobileNetV2_x1_5":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x1_5_pretrained.pdparams",
"MobileNetV2_x2_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/MobileNetV2_x2_0_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
@ -155,10 +163,10 @@ class InvresiBlocks(nn.Layer):
class MobileNet(nn.Layer):
def __init__(self, class_dim=1000, scale=1.0, prefix_name=""):
def __init__(self, class_num=1000, scale=1.0, prefix_name=""):
super(MobileNet, self).__init__()
self.scale = scale
self.class_dim = class_dim
self.class_num = class_num
bottleneck_params_list = [
(1, 16, 1, 1),
@ -209,7 +217,7 @@ class MobileNet(nn.Layer):
self.out = Linear(
self.out_c,
class_dim,
class_num,
weight_attr=ParamAttr(name=prefix_name + "fc10_weights"),
bias_attr=ParamAttr(name=prefix_name + "fc10_offset"))
@ -222,8 +230,8 @@ class MobileNet(nn.Layer):
y = paddle.flatten(y, start_axis=1, stop_axis=-1)
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -235,39 +243,45 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def MobileNetV2_x0_25(pretrained=False, use_ssld=False, **kwargs):
model = MobileNet(scale=0.25, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MobileNetV2_x0_25"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MobileNetV2_x0_25"], use_ssld=use_ssld)
return model
def MobileNetV2_x0_5(pretrained=False, use_ssld=False, **kwargs):
model = MobileNet(scale=0.5, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MobileNetV2_x0_5"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MobileNetV2_x0_5"], use_ssld=use_ssld)
return model
def MobileNetV2_x0_75(pretrained=False, use_ssld=False, **kwargs):
model = MobileNet(scale=0.75, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MobileNetV2_x0_75"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MobileNetV2_x0_75"], use_ssld=use_ssld)
return model
def MobileNetV2(pretrained=False, use_ssld=False, **kwargs):
model = MobileNet(scale=1.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MobileNetV2"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MobileNetV2"], use_ssld=use_ssld)
return model
def MobileNetV2_x1_5(pretrained=False, use_ssld=False, **kwargs):
model = MobileNet(scale=1.5, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MobileNetV2_x1_5"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MobileNetV2_x1_5"], use_ssld=use_ssld)
return model
def MobileNetV2_x2_0(pretrained=False, use_ssld=False, **kwargs):
model = MobileNet(scale=2.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["MobileNetV2_x2_0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["MobileNetV2_x2_0"], use_ssld=use_ssld)
return model

75
ppcls/arch/backbone/model_zoo/rednet.py
View File

@ -19,7 +19,6 @@ from paddle.vision.models import resnet
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"RedNet26":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RedNet26_pretrained.pdparams",
@ -33,7 +32,6 @@ MODEL_URLS = {
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RedNet152_pretrained.pdparams"
}
__all__ = MODEL_URLS.keys()
@ -51,50 +49,53 @@ class Involution(nn.Layer):
in_channels=channels,
out_channels=channels // reduction_ratio,
kernel_size=1,
bias_attr=False
)),
bias_attr=False)),
('bn', nn.BatchNorm2D(channels // reduction_ratio)),
('activate', nn.ReLU())
)
self.conv2 = nn.Sequential(
('conv', nn.Conv2D(
in_channels=channels // reduction_ratio,
out_channels=kernel_size**2 * self.groups,
kernel_size=1,
stride=1
))
)
('activate', nn.ReLU()))
self.conv2 = nn.Sequential(('conv', nn.Conv2D(
in_channels=channels // reduction_ratio,
out_channels=kernel_size**2 * self.groups,
kernel_size=1,
stride=1)))
if stride > 1:
self.avgpool = nn.AvgPool2D(stride, stride)
def forward(self, x):
weight = self.conv2(self.conv1(x if self.stride == 1 else self.avgpool(x)))
weight = self.conv2(
self.conv1(x if self.stride == 1 else self.avgpool(x)))
b, c, h, w = weight.shape
weight = weight.reshape((b, self.groups, self.kernel_size**2, h, w)).unsqueeze(2)
weight = weight.reshape(
(b, self.groups, self.kernel_size**2, h, w)).unsqueeze(2)
out = nn.functional.unfold(x, self.kernel_size, self.stride, (self.kernel_size-1)//2, 1)
out = out.reshape((b, self.groups, self.group_channels, self.kernel_size**2, h, w))
out = nn.functional.unfold(x, self.kernel_size, self.stride,
(self.kernel_size - 1) // 2, 1)
out = out.reshape(
(b, self.groups, self.group_channels, self.kernel_size**2, h, w))
out = (weight * out).sum(axis=3).reshape((b, self.channels, h, w))
return out
class BottleneckBlock(resnet.BottleneckBlock):
def __init__(self, inplanes, planes, stride=1, downsample=None,
groups=1, base_width=64, dilation=1, norm_layer=None):
super(BottleneckBlock, self).__init__(
inplanes, planes, stride, downsample,
groups, base_width, dilation, norm_layer
)
def __init__(self,
inplanes,
planes,
stride=1,
downsample=None,
groups=1,
base_width=64,
dilation=1,
norm_layer=None):
super(BottleneckBlock, self).__init__(inplanes, planes, stride,
downsample, groups, base_width,
dilation, norm_layer)
width = int(planes * (base_width / 64.)) * groups
self.conv2 = Involution(width, 7, stride)
class RedNet(resnet.ResNet):
def __init__(self, block, depth, class_dim=1000, with_pool=True):
def __init__(self, block, depth, class_num=1000, with_pool=True):
super(RedNet, self).__init__(
block=block, depth=50,
num_classes=class_dim, with_pool=with_pool
)
block=block, depth=50, num_classes=class_num, with_pool=with_pool)
layer_cfg = {
26: [1, 2, 4, 1],
38: [2, 3, 5, 2],
@ -108,7 +109,7 @@ class RedNet(resnet.ResNet):
self.bn1 = None
self.relu = None
self.inplanes = 64
self.class_dim = class_dim
self.class_num = class_num
self.stem = nn.Sequential(
nn.Sequential(
('conv', nn.Conv2D(
@ -117,11 +118,9 @@ class RedNet(resnet.ResNet):
kernel_size=3,
stride=2,
padding=1,
bias_attr=False
)),
bias_attr=False)),
('bn', nn.BatchNorm2D(self.inplanes // 2)),
('activate', nn.ReLU())
),
('activate', nn.ReLU())),
Involution(self.inplanes // 2, 3, 1),
nn.BatchNorm2D(self.inplanes // 2),
nn.ReLU(),
@ -132,12 +131,8 @@ class RedNet(resnet.ResNet):
kernel_size=3,
stride=1,
padding=1,
bias_attr=False
)),
('bn', nn.BatchNorm2D(self.inplanes)),
('activate', nn.ReLU())
)
)
bias_attr=False)), ('bn', nn.BatchNorm2D(self.inplanes)),
('activate', nn.ReLU())))
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
@ -156,7 +151,7 @@ class RedNet(resnet.ResNet):
if self.with_pool:
x = self.avgpool(x)
if self.class_dim > 0:
if self.class_num > 0:
x = paddle.flatten(x, 1)
x = self.fc(x)

58
ppcls/arch/backbone/model_zoo/regnet.py
View File

@ -28,13 +28,20 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"RegNetX_200MF": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetX_200MF_pretrained.pdparams",
"RegNetX_4GF": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetX_4GF_pretrained.pdparams",
"RegNetX_32GF": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetX_32GF_pretrained.pdparams",
"RegNetY_200MF": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetY_200MF_pretrained.pdparams",
"RegNetY_4GF": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetY_4GF_pretrained.pdparams",
"RegNetY_32GF": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetY_32GF_pretrained.pdparams",
}
MODEL_URLS = {
"RegNetX_200MF":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetX_200MF_pretrained.pdparams",
"RegNetX_4GF":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetX_4GF_pretrained.pdparams",
"RegNetX_32GF":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetX_32GF_pretrained.pdparams",
"RegNetY_200MF":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetY_200MF_pretrained.pdparams",
"RegNetY_4GF":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetY_4GF_pretrained.pdparams",
"RegNetY_32GF":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RegNetY_32GF_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -235,7 +242,7 @@ class RegNet(nn.Layer):
bot_mul,
q=8,
se_on=False,
class_dim=1000):
class_num=1000):
super(RegNet, self).__init__()
# Generate RegNet ws per block
@ -301,7 +308,7 @@ class RegNet(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_0.w_0"),
bias_attr=ParamAttr(name="fc_0.b_0"))
@ -315,7 +322,7 @@ class RegNet(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -327,12 +334,20 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def RegNetX_200MF(pretrained=False, use_ssld=False, **kwargs):
model = RegNet(
w_a=36.44, w_0=24, w_m=2.49, d=13, group_w=8, bot_mul=1.0, q=8, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RegNetX_200MF"], use_ssld=use_ssld)
w_a=36.44,
w_0=24,
w_m=2.49,
d=13,
group_w=8,
bot_mul=1.0,
q=8,
**kwargs)
_load_pretrained(
pretrained, model, MODEL_URLS["RegNetX_200MF"], use_ssld=use_ssld)
return model
@ -346,7 +361,8 @@ def RegNetX_4GF(pretrained=False, use_ssld=False, **kwargs):
bot_mul=1.0,
q=8,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RegNetX_4GF"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RegNetX_4GF"], use_ssld=use_ssld)
return model
@ -360,7 +376,8 @@ def RegNetX_32GF(pretrained=False, use_ssld=False, **kwargs):
bot_mul=1.0,
q=8,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
return model
@ -375,7 +392,8 @@ def RegNetY_200MF(pretrained=False, use_ssld=False, **kwargs):
q=8,
se_on=True,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
return model
@ -390,7 +408,8 @@ def RegNetY_4GF(pretrained=False, use_ssld=False, **kwargs):
q=8,
se_on=True,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
return model
@ -405,5 +424,6 @@ def RegNetY_32GF(pretrained=False, use_ssld=False, **kwargs):
q=8,
se_on=True,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RegNetX_32GF"], use_ssld=use_ssld)
return model

88
ppcls/arch/backbone/model_zoo/repvgg.py
View File

@ -4,24 +4,37 @@ import numpy as np
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"RepVGG_A0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_A0_pretrained.pdparams",
"RepVGG_A1": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_A1_pretrained.pdparams",
"RepVGG_A2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_A2_pretrained.pdparams",
"RepVGG_B0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B0_pretrained.pdparams",
"RepVGG_B1": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B1_pretrained.pdparams",
"RepVGG_B2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B2_pretrained.pdparams",
"RepVGG_B3": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B3_pretrained.pdparams",
"RepVGG_B1g2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B1g2_pretrained.pdparams",
"RepVGG_B1g4": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B1g4_pretrained.pdparams",
"RepVGG_B2g2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B2g2_pretrained.pdparams",
"RepVGG_B2g4": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B2g4_pretrained.pdparams",
"RepVGG_B3g2": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B3g2_pretrained.pdparams",
"RepVGG_B3g4": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B3g4_pretrained.pdparams",
}
MODEL_URLS = {
"RepVGG_A0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_A0_pretrained.pdparams",
"RepVGG_A1":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_A1_pretrained.pdparams",
"RepVGG_A2":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_A2_pretrained.pdparams",
"RepVGG_B0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B0_pretrained.pdparams",
"RepVGG_B1":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B1_pretrained.pdparams",
"RepVGG_B2":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B2_pretrained.pdparams",
"RepVGG_B3":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B3_pretrained.pdparams",
"RepVGG_B1g2":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B1g2_pretrained.pdparams",
"RepVGG_B1g4":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B1g4_pretrained.pdparams",
"RepVGG_B2g2":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B2g2_pretrained.pdparams",
"RepVGG_B2g4":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B2g4_pretrained.pdparams",
"RepVGG_B3g2":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B3g2_pretrained.pdparams",
"RepVGG_B3g4":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/RepVGG_B3g4_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
optional_groupwise_layers = [2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26]
g2_map = {l: 2 for l in optional_groupwise_layers}
g4_map = {l: 4 for l in optional_groupwise_layers}
@ -174,7 +187,7 @@ class RepVGG(nn.Layer):
num_blocks,
width_multiplier=None,
override_groups_map=None,
class_dim=1000):
class_num=1000):
super(RepVGG, self).__init__()
assert len(width_multiplier) == 4
@ -200,7 +213,7 @@ class RepVGG(nn.Layer):
self.stage4 = self._make_stage(
int(512 * width_multiplier[3]), num_blocks[3], stride=2)
self.gap = nn.AdaptiveAvgPool2D(output_size=1)
self.linear = nn.Linear(int(512 * width_multiplier[3]), class_dim)
self.linear = nn.Linear(int(512 * width_multiplier[3]), class_num)
def _make_stage(self, planes, num_blocks, stride):
strides = [stride] + [1] * (num_blocks - 1)
@ -248,7 +261,7 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def RepVGG_A0(pretrained=False, use_ssld=False, **kwargs):
model = RepVGG(
@ -256,7 +269,8 @@ def RepVGG_A0(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[0.75, 0.75, 0.75, 2.5],
override_groups_map=None,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_A0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_A0"], use_ssld=use_ssld)
return model
@ -266,7 +280,8 @@ def RepVGG_A1(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[1, 1, 1, 2.5],
override_groups_map=None,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_A1"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_A1"], use_ssld=use_ssld)
return model
@ -276,7 +291,8 @@ def RepVGG_A2(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[1.5, 1.5, 1.5, 2.75],
override_groups_map=None,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_A2"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_A2"], use_ssld=use_ssld)
return model
@ -286,7 +302,8 @@ def RepVGG_B0(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[1, 1, 1, 2.5],
override_groups_map=None,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B0"], use_ssld=use_ssld)
return model
@ -296,7 +313,8 @@ def RepVGG_B1(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[2, 2, 2, 4],
override_groups_map=None,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B1"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B1"], use_ssld=use_ssld)
return model
@ -306,7 +324,8 @@ def RepVGG_B1g2(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[2, 2, 2, 4],
override_groups_map=g2_map,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B1g2"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B1g2"], use_ssld=use_ssld)
return model
@ -316,7 +335,8 @@ def RepVGG_B1g4(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[2, 2, 2, 4],
override_groups_map=g4_map,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B1g4"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B1g4"], use_ssld=use_ssld)
return model
@ -326,7 +346,8 @@ def RepVGG_B2(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[2.5, 2.5, 2.5, 5],
override_groups_map=None,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B2"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B2"], use_ssld=use_ssld)
return model
@ -336,7 +357,8 @@ def RepVGG_B2g2(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[2.5, 2.5, 2.5, 5],
override_groups_map=g2_map,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B2g2"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B2g2"], use_ssld=use_ssld)
return model
@ -346,7 +368,8 @@ def RepVGG_B2g4(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[2.5, 2.5, 2.5, 5],
override_groups_map=g4_map,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B2g4"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B2g4"], use_ssld=use_ssld)
return model
@ -356,7 +379,8 @@ def RepVGG_B3(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[3, 3, 3, 5],
override_groups_map=None,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B3"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B3"], use_ssld=use_ssld)
return model
@ -366,7 +390,8 @@ def RepVGG_B3g2(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[3, 3, 3, 5],
override_groups_map=g2_map,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B3g2"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B3g2"], use_ssld=use_ssld)
return model
@ -376,5 +401,6 @@ def RepVGG_B3g4(pretrained=False, use_ssld=False, **kwargs):
width_multiplier=[3, 3, 3, 5],
override_groups_map=g4_map,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["RepVGG_B3g4"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["RepVGG_B3g4"], use_ssld=use_ssld)
return model

23
ppcls/arch/backbone/model_zoo/res2net.py
View File

@ -29,9 +29,12 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"Res2Net50_26w_4s": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net50_26w_4s_pretrained.pdparams",
"Res2Net50_14w_8s": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net50_14w_8s_pretrained.pdparams",
}
MODEL_URLS = {
"Res2Net50_26w_4s":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net50_26w_4s_pretrained.pdparams",
"Res2Net50_14w_8s":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net50_14w_8s_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -151,7 +154,7 @@ class BottleneckBlock(nn.Layer):
class Res2Net(nn.Layer):
def __init__(self, layers=50, scales=4, width=26, class_dim=1000):
def __init__(self, layers=50, scales=4, width=26, class_num=1000):
super(Res2Net, self).__init__()
self.layers = layers
@ -218,7 +221,7 @@ class Res2Net(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_weights"),
bias_attr=ParamAttr(name="fc_offset"))
@ -245,15 +248,17 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def Res2Net50_26w_4s(pretrained=False, use_ssld=False, **kwargs):
model = Res2Net(layers=50, scales=4, width=26, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Res2Net50_26w_4s"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["Res2Net50_26w_4s"], use_ssld=use_ssld)
return model
def Res2Net50_14w_8s(pretrained=False, use_ssld=False, **kwargs):
model = Res2Net(layers=50, scales=8, width=14, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Res2Net50_14w_8s"], use_ssld=use_ssld)
return model
_load_pretrained(
pretrained, model, MODEL_URLS["Res2Net50_14w_8s"], use_ssld=use_ssld)
return model

36
ppcls/arch/backbone/model_zoo/res2net_vd.py
View File

@ -29,10 +29,14 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"Res2Net50_vd_26w_4s": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net50_vd_26w_4s_pretrained.pdparams",
"Res2Net101_vd_26w_4s": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net101_vd_26w_4s_pretrained.pdparams",
"Res2Net200_vd_26w_4s": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net200_vd_26w_4s_pretrained.pdparams",
}
MODEL_URLS = {
"Res2Net50_vd_26w_4s":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net50_vd_26w_4s_pretrained.pdparams",
"Res2Net101_vd_26w_4s":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net101_vd_26w_4s_pretrained.pdparams",
"Res2Net200_vd_26w_4s":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Res2Net200_vd_26w_4s_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -159,7 +163,7 @@ class BottleneckBlock(nn.Layer):
class Res2Net_vd(nn.Layer):
def __init__(self, layers=50, scales=4, width=26, class_dim=1000):
def __init__(self, layers=50, scales=4, width=26, class_num=1000):
super(Res2Net_vd, self).__init__()
self.layers = layers
@ -240,7 +244,7 @@ class Res2Net_vd(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_weights"),
bias_attr=ParamAttr(name="fc_offset"))
@ -273,17 +277,29 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def Res2Net50_vd_26w_4s(pretrained=False, use_ssld=False, **kwargs):
model = Res2Net_vd(layers=50, scales=4, width=26, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Res2Net50_vd_26w_4s"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["Res2Net50_vd_26w_4s"],
use_ssld=use_ssld)
return model
def Res2Net101_vd_26w_4s(pretrained=False, use_ssld=False, **kwargs):
model = Res2Net_vd(layers=101, scales=4, width=26, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Res2Net101_vd_26w_4s"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["Res2Net101_vd_26w_4s"],
use_ssld=use_ssld)
return model
def Res2Net200_vd_26w_4s(pretrained=False, use_ssld=False, **kwargs):
model = Res2Net_vd(layers=200, scales=4, width=26, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Res2Net200_vd_26w_4s"], use_ssld=use_ssld)
return model
_load_pretrained(
pretrained,
model,
MODEL_URLS["Res2Net200_vd_26w_4s"],
use_ssld=use_ssld)
return model

34
ppcls/arch/backbone/model_zoo/resnest.py
View File

@ -29,10 +29,14 @@ from paddle.regularizer import L2Decay
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"ResNeSt50_fast_1s1x64d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeSt50_fast_1s1x64d_pretrained.pdparams",
"ResNeSt50": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeSt50_pretrained.pdparams",
"ResNeSt101": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeSt101_pretrained.pdparams",
}
MODEL_URLS = {
"ResNeSt50_fast_1s1x64d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeSt50_fast_1s1x64d_pretrained.pdparams",
"ResNeSt50":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeSt50_pretrained.pdparams",
"ResNeSt101":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeSt101_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -453,7 +457,7 @@ class ResNeSt(nn.Layer):
avd_first=False,
final_drop=0.0,
last_gamma=False,
class_dim=1000):
class_num=1000):
super(ResNeSt, self).__init__()
self.cardinality = groups
@ -643,7 +647,7 @@ class ResNeSt(nn.Layer):
self.out = Linear(
self.out_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=nn.initializer.Uniform(-stdv, stdv),
name="fc_weights"),
@ -663,7 +667,7 @@ class ResNeSt(nn.Layer):
x = self.out(x)
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -675,8 +679,8 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def ResNeSt50_fast_1s1x64d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeSt(
layers=[3, 4, 6, 3],
@ -690,7 +694,11 @@ def ResNeSt50_fast_1s1x64d(pretrained=False, use_ssld=False, **kwargs):
avd_first=True,
final_drop=0.0,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeSt50_fast_1s1x64d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeSt50_fast_1s1x64d"],
use_ssld=use_ssld)
return model
@ -707,7 +715,8 @@ def ResNeSt50(pretrained=False, use_ssld=False, **kwargs):
avd_first=False,
final_drop=0.0,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeSt50"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeSt50"], use_ssld=use_ssld)
return model
@ -724,5 +733,6 @@ def ResNeSt101(pretrained=False, use_ssld=False, **kwargs):
avd_first=False,
final_drop=0.0,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeSt101"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeSt101"], use_ssld=use_ssld)
return model

17
ppcls/arch/backbone/model_zoo/resnet_vc.py
View File

@ -30,8 +30,9 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"ResNet50_vc": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vc_pretrained.pdparams",
}
"ResNet50_vc":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNet50_vc_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -177,7 +178,7 @@ class BasicBlock(nn.Layer):
class ResNet_vc(nn.Layer):
def __init__(self, layers=50, class_dim=1000):
def __init__(self, layers=50, class_num=1000):
super(ResNet_vc, self).__init__()
self.layers = layers
@ -270,7 +271,7 @@ class ResNet_vc(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_0.w_0"),
bias_attr=ParamAttr(name="fc_0.b_0"))
@ -287,7 +288,7 @@ class ResNet_vc(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -300,9 +301,9 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
"pretrained type is not available. Please use `string` or `boolean` type."
)
def ResNet50_vc(pretrained=False, use_ssld=False, **kwargs):
model = ResNet_vc(layers=50, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNet50_vc"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNet50_vc"], use_ssld=use_ssld)
return model

52
ppcls/arch/backbone/model_zoo/resnext.py
View File

@ -30,13 +30,19 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"ResNeXt50_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_32x4d_pretrained.pdparams",
"ResNeXt50_64x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_64x4d_pretrained.pdparams",
"ResNeXt101_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x4d_pretrained.pdparams",
"ResNeXt101_64x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_64x4d_pretrained.pdparams",
"ResNeXt152_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_32x4d_pretrained.pdparams",
"ResNeXt152_64x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_64x4d_pretrained.pdparams",
}
"ResNeXt50_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_32x4d_pretrained.pdparams",
"ResNeXt50_64x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_64x4d_pretrained.pdparams",
"ResNeXt101_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x4d_pretrained.pdparams",
"ResNeXt101_64x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_64x4d_pretrained.pdparams",
"ResNeXt152_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_32x4d_pretrained.pdparams",
"ResNeXt152_64x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_64x4d_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -143,7 +149,12 @@ class BottleneckBlock(nn.Layer):
class ResNeXt(nn.Layer):
def __init__(self, layers=50, class_dim=1000, cardinality=32, input_image_channel=3, data_format="NCHW"):
def __init__(self,
layers=50,
class_num=1000,
cardinality=32,
input_image_channel=3,
data_format="NCHW"):
super(ResNeXt, self).__init__()
self.layers = layers
@ -176,7 +187,8 @@ class ResNeXt(nn.Layer):
act='relu',
name="res_conv1",
data_format=self.data_format)
self.pool2d_max = MaxPool2D(kernel_size=3, stride=2, padding=1, data_format=self.data_format)
self.pool2d_max = MaxPool2D(
kernel_size=3, stride=2, padding=1, data_format=self.data_format)
self.block_list = []
for block in range(len(depth)):
@ -211,7 +223,7 @@ class ResNeXt(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_weights"),
bias_attr=ParamAttr(name="fc_offset"))
@ -230,7 +242,7 @@ class ResNeXt(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -246,35 +258,41 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def ResNeXt50_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=50, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt50_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt50_32x4d"], use_ssld=use_ssld)
return model
def ResNeXt50_64x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=50, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt50_64x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt50_64x4d"], use_ssld=use_ssld)
return model
def ResNeXt101_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=101, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt101_32x4d"], use_ssld=use_ssld)
return model
def ResNeXt101_64x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=101, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_64x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt101_64x4d"], use_ssld=use_ssld)
return model
def ResNeXt152_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=152, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt152_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt152_32x4d"], use_ssld=use_ssld)
return model
def ResNeXt152_64x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=152, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt152_64x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt152_64x4d"], use_ssld=use_ssld)
return model

48
ppcls/arch/backbone/model_zoo/resnext101_wsl.py
View File

@ -9,17 +9,19 @@ from paddle.nn.initializer import Uniform
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"ResNeXt101_32x8d_wsl": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x8d_wsl_pretrained.pdparams",
"ResNeXt101_32x16d_wsl": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x816_wsl_pretrained.pdparams",
"ResNeXt101_32x32d_wsl": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x32d_wsl_pretrained.pdparams",
"ResNeXt101_32x48d_wsl": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x48d_wsl_pretrained.pdparams",
}
"ResNeXt101_32x8d_wsl":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x8d_wsl_pretrained.pdparams",
"ResNeXt101_32x16d_wsl":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x816_wsl_pretrained.pdparams",
"ResNeXt101_32x32d_wsl":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x32d_wsl_pretrained.pdparams",
"ResNeXt101_32x48d_wsl":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_32x48d_wsl_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
class ConvBNLayer(nn.Layer):
def __init__(self,
input_channels,
@ -128,10 +130,10 @@ class BottleneckBlock(nn.Layer):
class ResNeXt101WSL(nn.Layer):
def __init__(self, layers=101, cardinality=32, width=48, class_dim=1000):
def __init__(self, layers=101, cardinality=32, width=48, class_num=1000):
super(ResNeXt101WSL, self).__init__()
self.class_dim = class_dim
self.class_num = class_num
self.layers = layers
self.cardinality = cardinality
@ -384,7 +386,7 @@ class ResNeXt101WSL(nn.Layer):
self._avg_pool = AdaptiveAvgPool2D(1)
self._out = Linear(
num_filters[3] // (width // 8),
class_dim,
class_num,
weight_attr=ParamAttr(name="fc.weight"),
bias_attr=ParamAttr(name="fc.bias"))
@ -434,7 +436,7 @@ class ResNeXt101WSL(nn.Layer):
x = self._out(x)
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -450,23 +452,39 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def ResNeXt101_32x8d_wsl(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt101WSL(cardinality=32, width=8, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_32x8d_wsl"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt101_32x8d_wsl"],
use_ssld=use_ssld)
return model
def ResNeXt101_32x16d_wsl(**args):
model = ResNeXt101WSL(cardinality=32, width=16, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_32x16d_ws"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt101_32x16d_ws"],
use_ssld=use_ssld)
return model
def ResNeXt101_32x32d_wsl(**args):
model = ResNeXt101WSL(cardinality=32, width=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_32x32d_wsl"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt101_32x32d_wsl"],
use_ssld=use_ssld)
return model
def ResNeXt101_32x48d_wsl(**args):
model = ResNeXt101WSL(cardinality=32, width=48, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_32x48d_wsl"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt101_32x48d_wsl"],
use_ssld=use_ssld)
return model

56
ppcls/arch/backbone/model_zoo/resnext_vd.py
View File

@ -30,16 +30,23 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"ResNeXt50_vd_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_vd_32x4d_pretrained.pdparams",
"ResNeXt50_vd_64x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_vd_64x4d_pretrained.pdparams",
"ResNeXt101_vd_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_vd_32x4d_pretrained.pdparams",
"ResNeXt101_vd_64x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_vd_64x4d_pretrained.pdparams",
"ResNeXt152_vd_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_vd_32x4d_pretrained.pdparams",
"ResNeXt152_vd_64x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_vd_64x4d_pretrained.pdparams",
}
"ResNeXt50_vd_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_vd_32x4d_pretrained.pdparams",
"ResNeXt50_vd_64x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt50_vd_64x4d_pretrained.pdparams",
"ResNeXt101_vd_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_vd_32x4d_pretrained.pdparams",
"ResNeXt101_vd_64x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt101_vd_64x4d_pretrained.pdparams",
"ResNeXt152_vd_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_vd_32x4d_pretrained.pdparams",
"ResNeXt152_vd_64x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ResNeXt152_vd_64x4d_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
class ConvBNLayer(nn.Layer):
def __init__(
self,
@ -145,7 +152,7 @@ class BottleneckBlock(nn.Layer):
class ResNeXt(nn.Layer):
def __init__(self, layers=50, class_dim=1000, cardinality=32):
def __init__(self, layers=50, class_num=1000, cardinality=32):
super(ResNeXt, self).__init__()
self.layers = layers
@ -225,7 +232,7 @@ class ResNeXt(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc_weights"),
bias_attr=ParamAttr(name="fc_offset"))
@ -242,6 +249,7 @@ class ResNeXt(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -257,35 +265,53 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def ResNeXt50_vd_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=50, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt50_vd_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt50_vd_32x4d"], use_ssld=use_ssld)
return model
def ResNeXt50_vd_64x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=50, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt50_vd_64x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ResNeXt50_vd_64x4d"], use_ssld=use_ssld)
return model
def ResNeXt101_vd_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=101, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_vd_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt101_vd_32x4d"],
use_ssld=use_ssld)
return model
def ResNeXt101_vd_64x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=101, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt101_vd_64x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt101_vd_64x4d"],
use_ssld=use_ssld)
return model
def ResNeXt152_vd_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=152, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt152_vd_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt152_vd_32x4d"],
use_ssld=use_ssld)
return model
def ResNeXt152_vd_64x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=152, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ResNeXt152_vd_64x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["ResNeXt152_vd_64x4d"],
use_ssld=use_ssld)
return model

42
ppcls/arch/backbone/model_zoo/rexnet.py
View File

@ -25,12 +25,17 @@ from math import ceil
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"ReXNet_1_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_1_0_pretrained.pdparams",
"ReXNet_1_3": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_1_3_pretrained.pdparams",
"ReXNet_1_5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_1_5_32x4d_pretrained.pdparams",
"ReXNet_2_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_2_0_pretrained.pdparams",
"ReXNet_3_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_3_0_pretrained.pdparams",
}
"ReXNet_1_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_1_0_pretrained.pdparams",
"ReXNet_1_3":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_1_3_pretrained.pdparams",
"ReXNet_1_5":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_1_5_32x4d_pretrained.pdparams",
"ReXNet_2_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_2_0_pretrained.pdparams",
"ReXNet_3_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ReXNet_3_0_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -154,7 +159,7 @@ class ReXNetV1(nn.Layer):
final_ch=180,
width_mult=1.0,
depth_mult=1.0,
class_dim=1000,
class_num=1000,
use_se=True,
se_ratio=12,
dropout_ratio=0.2,
@ -220,7 +225,7 @@ class ReXNetV1(nn.Layer):
self.output = nn.Sequential(
nn.Dropout(dropout_ratio),
nn.Conv2D(
pen_channels, class_dim, 1, bias_attr=True))
pen_channels, class_num, 1, bias_attr=True))
def forward(self, x):
x = self.features(x)
@ -239,33 +244,38 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def ReXNet_1_0(pretrained=False, use_ssld=False, **kwargs):
model = ReXNetV1(width_mult=1.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ReXNet_1_0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ReXNet_1_0"], use_ssld=use_ssld)
return model
def ReXNet_1_3(pretrained=False, use_ssld=False, **kwargs):
model = ReXNetV1(width_mult=1.3, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ReXNet_1_3"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ReXNet_1_3"], use_ssld=use_ssld)
return model
def ReXNet_1_5(pretrained=False, use_ssld=False, **kwargs):
model = ReXNetV1(width_mult=1.5, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ReXNet_1_5"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ReXNet_1_5"], use_ssld=use_ssld)
return model
def ReXNet_2_0(pretrained=False, use_ssld=False, **kwargs):
model = ReXNetV1(width_mult=2.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ReXNet_2_0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ReXNet_2_0"], use_ssld=use_ssld)
return model
def ReXNet_3_0(pretrained=False, use_ssld=False, **kwargs):
model = ReXNetV1(width_mult=3.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ReXNet_3_0"], use_ssld=use_ssld)
return model
_load_pretrained(
pretrained, model, MODEL_URLS["ReXNet_3_0"], use_ssld=use_ssld)
return model

29
ppcls/arch/backbone/model_zoo/se_resnet_vd.py
View File

@ -29,11 +29,13 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"SE_ResNet18_vd": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNet18_vd_pretrained.pdparams",
"SE_ResNet34_vd": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNet34_vd_pretrained.pdparams",
"SE_ResNet50_vd": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNet50_vd_pretrained.pdparams",
}
"SE_ResNet18_vd":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNet18_vd_pretrained.pdparams",
"SE_ResNet34_vd":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNet34_vd_pretrained.pdparams",
"SE_ResNet50_vd":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNet50_vd_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -240,7 +242,7 @@ class SELayer(nn.Layer):
class SE_ResNet_vd(nn.Layer):
def __init__(self, layers=50, class_dim=1000):
def __init__(self, layers=50, class_num=1000):
super(SE_ResNet_vd, self).__init__()
self.layers = layers
@ -336,7 +338,7 @@ class SE_ResNet_vd(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc6_weights"),
bias_attr=ParamAttr(name="fc6_offset"))
@ -353,7 +355,7 @@ class SE_ResNet_vd(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -365,21 +367,24 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def SE_ResNet18_vd(pretrained=False, use_ssld=False, **kwargs):
model = SE_ResNet_vd(layers=18, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNet18_vd"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["SE_ResNet18_vd"], use_ssld=use_ssld)
return model
def SE_ResNet34_vd(pretrained=False, use_ssld=False, **kwargs):
model = SE_ResNet_vd(layers=34, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNet34_vd"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["SE_ResNet34_vd"], use_ssld=use_ssld)
return model
def SE_ResNet50_vd(pretrained=False, use_ssld=False, **kwargs):
model = SE_ResNet_vd(layers=50, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNet50_vd"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["SE_ResNet50_vd"], use_ssld=use_ssld)
return model

52
ppcls/arch/backbone/model_zoo/se_resnext.py
View File

@ -30,11 +30,13 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"SE_ResNeXt50_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt50_32x4d_pretrained.pdparams",
"SE_ResNeXt101_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt101_32x4d_pretrained.pdparams",
"SE_ResNeXt152_64x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt152_64x4d_pretrained.pdparams",
}
"SE_ResNeXt50_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt50_32x4d_pretrained.pdparams",
"SE_ResNeXt101_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt101_32x4d_pretrained.pdparams",
"SE_ResNeXt152_64x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt152_64x4d_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -148,7 +150,12 @@ class BottleneckBlock(nn.Layer):
class SELayer(nn.Layer):
def __init__(self, num_channels, num_filters, reduction_ratio, name=None, data_format="NCHW"):
def __init__(self,
num_channels,
num_filters,
reduction_ratio,
name=None,
data_format="NCHW"):
super(SELayer, self).__init__()
self.data_format = data_format
@ -193,7 +200,12 @@ class SELayer(nn.Layer):
class ResNeXt(nn.Layer):
def __init__(self, layers=50, class_dim=1000, cardinality=32, input_image_channel=3, data_format="NCHW"):
def __init__(self,
layers=50,
class_num=1000,
cardinality=32,
input_image_channel=3,
data_format="NCHW"):
super(ResNeXt, self).__init__()
self.layers = layers
@ -254,7 +266,8 @@ class ResNeXt(nn.Layer):
name="conv3",
data_format=self.data_format)
self.pool2d_max = MaxPool2D(kernel_size=3, stride=2, padding=1, data_format=self.data_format)
self.pool2d_max = MaxPool2D(
kernel_size=3, stride=2, padding=1, data_format=self.data_format)
self.block_list = []
n = 1 if layers == 50 or layers == 101 else 3
@ -286,13 +299,13 @@ class ResNeXt(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc6_weights"),
bias_attr=ParamAttr(name="fc6_offset"))
def forward(self, inputs):
with paddle.static.amp.fp16_guard():
with paddle.static.amp.fp16_guard():
if self.data_format == "NHWC":
inputs = paddle.tensor.transpose(inputs, [0, 2, 3, 1])
inputs.stop_gradient = True
@ -310,7 +323,7 @@ class ResNeXt(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -322,21 +335,30 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def SE_ResNeXt50_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=50, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNeXt50_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["SE_ResNeXt50_32x4d"], use_ssld=use_ssld)
return model
def SE_ResNeXt101_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=101, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNeXt101_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SE_ResNeXt101_32x4d"],
use_ssld=use_ssld)
return model
def SE_ResNeXt152_64x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=152, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNeXt152_64x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SE_ResNeXt152_64x4d"],
use_ssld=use_ssld)
return model

35
ppcls/arch/backbone/model_zoo/se_resnext_vd.py
View File

@ -30,11 +30,13 @@ import math
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"SE_ResNeXt50_vd_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt50_vd_32x4d_pretrained.pdparams",
"SE_ResNeXt50_vd_32x4d": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt50_vd_32x4d_pretrained.pdparams",
"SENet154_vd": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SENet154_vd_pretrained.pdparams",
}
"SE_ResNeXt50_vd_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt50_vd_32x4d_pretrained.pdparams",
"SE_ResNeXt50_vd_32x4d":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SE_ResNeXt50_vd_32x4d_pretrained.pdparams",
"SENet154_vd":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SENet154_vd_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -184,7 +186,7 @@ class SELayer(nn.Layer):
class ResNeXt(nn.Layer):
def __init__(self, layers=50, class_dim=1000, cardinality=32):
def __init__(self, layers=50, class_num=1000, cardinality=32):
super(ResNeXt, self).__init__()
self.layers = layers
@ -261,7 +263,7 @@ class ResNeXt(nn.Layer):
self.out = Linear(
self.pool2d_avg_channels,
class_dim,
class_num,
weight_attr=ParamAttr(
initializer=Uniform(-stdv, stdv), name="fc6_weights"),
bias_attr=ParamAttr(name="fc6_offset"))
@ -278,7 +280,7 @@ class ResNeXt(nn.Layer):
y = self.out(y)
return y
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -290,21 +292,30 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def SE_ResNeXt50_vd_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=50, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNeXt50_vd_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SE_ResNeXt50_vd_32x4d"],
use_ssld=use_ssld)
return model
def SE_ResNeXt101_vd_32x4d(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=101, cardinality=32, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SE_ResNeXt101_vd_32x4d"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SE_ResNeXt101_vd_32x4d"],
use_ssld=use_ssld)
return model
def SENet154_vd(pretrained=False, use_ssld=False, **kwargs):
model = ResNeXt(layers=152, cardinality=64, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SENet154_vd"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["SENet154_vd"], use_ssld=use_ssld)
return model

50
ppcls/arch/backbone/model_zoo/shufflenet_v2.py
View File

@ -25,14 +25,21 @@ from paddle.nn.functional import swish
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"ShuffleNetV2_x0_25": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x0_25_pretrained.pdparams",
"ShuffleNetV2_x0_33": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x0_33_pretrained.pdparams",
"ShuffleNetV2_x0_5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x0_5_pretrained.pdparams",
"ShuffleNetV2_x1_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x1_0_pretrained.pdparams",
"ShuffleNetV2_x1_5": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x1_5_pretrained.pdparams",
"ShuffleNetV2_x2_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x2_0_pretrained.pdparams",
"ShuffleNetV2_swish": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_swish_pretrained.pdparams"
}
"ShuffleNetV2_x0_25":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x0_25_pretrained.pdparams",
"ShuffleNetV2_x0_33":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x0_33_pretrained.pdparams",
"ShuffleNetV2_x0_5":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x0_5_pretrained.pdparams",
"ShuffleNetV2_x1_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x1_0_pretrained.pdparams",
"ShuffleNetV2_x1_5":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x1_5_pretrained.pdparams",
"ShuffleNetV2_x2_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_x2_0_pretrained.pdparams",
"ShuffleNetV2_swish":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ShuffleNetV2_swish_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
@ -207,10 +214,10 @@ class InvertedResidualDS(Layer):
class ShuffleNet(Layer):
def __init__(self, class_dim=1000, scale=1.0, act="relu"):
def __init__(self, class_num=1000, scale=1.0, act="relu"):
super(ShuffleNet, self).__init__()
self.scale = scale
self.class_dim = class_dim
self.class_num = class_num
stage_repeats = [4, 8, 4]
if scale == 0.25:
@ -277,7 +284,7 @@ class ShuffleNet(Layer):
# 5. fc
self._fc = Linear(
stage_out_channels[-1],
class_dim,
class_num,
weight_attr=ParamAttr(name='fc6_weights'),
bias_attr=ParamAttr(name='fc6_offset'))
@ -308,41 +315,48 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def ShuffleNetV2_x0_25(pretrained=False, use_ssld=False, **kwargs):
model = ShuffleNet(scale=0.25, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ShuffleNetV2_x0_25"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ShuffleNetV2_x0_25"], use_ssld=use_ssld)
return model
def ShuffleNetV2_x0_33(pretrained=False, use_ssld=False, **kwargs):
model = ShuffleNet(scale=0.33, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ShuffleNetV2_x0_33"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ShuffleNetV2_x0_33"], use_ssld=use_ssld)
return model
def ShuffleNetV2_x0_5(pretrained=False, use_ssld=False, **kwargs):
model = ShuffleNet(scale=0.5, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ShuffleNetV2_x0_5"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ShuffleNetV2_x0_5"], use_ssld=use_ssld)
return model
def ShuffleNetV2_x1_0(pretrained=False, use_ssld=False, **kwargs):
model = ShuffleNet(scale=1.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ShuffleNetV2_x1_0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ShuffleNetV2_x1_0"], use_ssld=use_ssld)
return model
def ShuffleNetV2_x1_5(pretrained=False, use_ssld=False, **kwargs):
model = ShuffleNet(scale=1.5, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ShuffleNetV2_x1_5"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ShuffleNetV2_x1_5"], use_ssld=use_ssld)
return model
def ShuffleNetV2_x2_0(pretrained=False, use_ssld=False, **kwargs):
model = ShuffleNet(scale=2.0, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ShuffleNetV2_x2_0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ShuffleNetV2_x2_0"], use_ssld=use_ssld)
return model
def ShuffleNetV2_swish(pretrained=False, use_ssld=False, **kwargs):
model = ShuffleNet(scale=1.0, act="swish", **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["ShuffleNetV2_swish"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["ShuffleNetV2_swish"], use_ssld=use_ssld)
return model

21
ppcls/arch/backbone/model_zoo/squeezenet.py
View File

@ -22,9 +22,11 @@ from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"SqueezeNet1_0": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SqueezeNet1_0_pretrained.pdparams",
"SqueezeNet1_1": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SqueezeNet1_1_pretrained.pdparams",
}
"SqueezeNet1_0":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SqueezeNet1_0_pretrained.pdparams",
"SqueezeNet1_1":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SqueezeNet1_1_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -78,7 +80,7 @@ class MakeFire(nn.Layer):
class SqueezeNet(nn.Layer):
def __init__(self, version, class_dim=1000):
def __init__(self, version, class_num=1000):
super(SqueezeNet, self).__init__()
self.version = version
@ -125,7 +127,7 @@ class SqueezeNet(nn.Layer):
self._drop = Dropout(p=0.5, mode="downscale_in_infer")
self._conv9 = Conv2D(
512,
class_dim,
class_num,
1,
weight_attr=ParamAttr(name="conv10_weights"),
bias_attr=ParamAttr(name="conv10_offset"))
@ -164,6 +166,7 @@ class SqueezeNet(nn.Layer):
x = paddle.squeeze(x, axis=[2, 3])
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -175,15 +178,17 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def SqueezeNet1_0(pretrained=False, use_ssld=False, **kwargs):
model = SqueezeNet(version="1.0", **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SqueezeNet1_0"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["SqueezeNet1_0"], use_ssld=use_ssld)
return model
def SqueezeNet1_1(pretrained=False, use_ssld=False, **kwargs):
model = SqueezeNet(version="1.1", **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SqueezeNet1_1"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["SqueezeNet1_1"], use_ssld=use_ssld)
return model

86
ppcls/arch/backbone/model_zoo/swin_transformer.py
View File

@ -24,13 +24,19 @@ from .vision_transformer import trunc_normal_, zeros_, ones_, to_2tuple, DropPat
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"SwinTransformer_tiny_patch4_window7_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_tiny_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_small_patch4_window7_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_small_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_base_patch4_window7_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_base_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_base_patch4_window12_384": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_base_patch4_window12_384_pretrained.pdparams",
"SwinTransformer_large_patch4_window7_224": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_large_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_large_patch4_window12_384": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_large_patch4_window12_384_pretrained.pdparams",
}
"SwinTransformer_tiny_patch4_window7_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_tiny_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_small_patch4_window7_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_small_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_base_patch4_window7_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_base_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_base_patch4_window12_384":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_base_patch4_window12_384_pretrained.pdparams",
"SwinTransformer_large_patch4_window7_224":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_large_patch4_window7_224_pretrained.pdparams",
"SwinTransformer_large_patch4_window12_384":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/SwinTransformer_large_patch4_window12_384_pretrained.pdparams",
}
__all__ = list(MODEL_URLS.keys())
@ -611,7 +617,7 @@ class SwinTransformer(nn.Layer):
img_size=224,
patch_size=4,
in_chans=3,
class_dim=1000,
class_num=1000,
embed_dim=96,
depths=[2, 2, 6, 2],
num_heads=[3, 6, 12, 24],
@ -629,7 +635,7 @@ class SwinTransformer(nn.Layer):
**kwargs):
super(SwinTransformer, self).__init__()
self.num_classes = num_classes = class_dim
self.num_classes = num_classes = class_num
self.num_layers = len(depths)
self.embed_dim = embed_dim
self.ape = ape
@ -729,7 +735,7 @@ class SwinTransformer(nn.Layer):
flops += self.num_features * self.num_classes
return flops
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -743,7 +749,9 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
)
def SwinTransformer_tiny_patch4_window7_224(pretrained=False, use_ssld=False, **kwargs):
def SwinTransformer_tiny_patch4_window7_224(pretrained=False,
use_ssld=False,
**kwargs):
model = SwinTransformer(
embed_dim=96,
depths=[2, 2, 6, 2],
@ -751,22 +759,34 @@ def SwinTransformer_tiny_patch4_window7_224(pretrained=False, use_ssld=False, **
window_size=7,
drop_path_rate=0.2,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SwinTransformer_tiny_patch4_window7_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SwinTransformer_tiny_patch4_window7_224"],
use_ssld=use_ssld)
return model
def SwinTransformer_small_patch4_window7_224(pretrained=False, use_ssld=False, **kwargs):
def SwinTransformer_small_patch4_window7_224(pretrained=False,
use_ssld=False,
**kwargs):
model = SwinTransformer(
embed_dim=96,
depths=[2, 2, 18, 2],
num_heads=[3, 6, 12, 24],
window_size=7,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SwinTransformer_small_patch4_window7_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SwinTransformer_small_patch4_window7_224"],
use_ssld=use_ssld)
return model
def SwinTransformer_base_patch4_window7_224(pretrained=False, use_ssld=False, **kwargs):
def SwinTransformer_base_patch4_window7_224(pretrained=False,
use_ssld=False,
**kwargs):
model = SwinTransformer(
embed_dim=128,
depths=[2, 2, 18, 2],
@ -774,11 +794,17 @@ def SwinTransformer_base_patch4_window7_224(pretrained=False, use_ssld=False, **
window_size=7,
drop_path_rate=0.5,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SwinTransformer_base_patch4_window7_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SwinTransformer_base_patch4_window7_224"],
use_ssld=use_ssld)
return model
def SwinTransformer_base_patch4_window12_384(pretrained=False, use_ssld=False, **kwargs):
def SwinTransformer_base_patch4_window12_384(pretrained=False,
use_ssld=False,
**kwargs):
model = SwinTransformer(
img_size=384,
embed_dim=128,
@ -787,22 +813,34 @@ def SwinTransformer_base_patch4_window12_384(pretrained=False, use_ssld=False, *
window_size=12,
drop_path_rate=0.5, # NOTE: do not appear in offical code
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SwinTransformer_base_patch4_window12_384"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SwinTransformer_base_patch4_window12_384"],
use_ssld=use_ssld)
return model
def SwinTransformer_large_patch4_window7_224(pretrained=False, use_ssld=False, **kwargs):
def SwinTransformer_large_patch4_window7_224(pretrained=False,
use_ssld=False,
**kwargs):
model = SwinTransformer(
embed_dim=192,
depths=[2, 2, 18, 2],
num_heads=[6, 12, 24, 48],
window_size=7,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SwinTransformer_large_patch4_window7_224"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SwinTransformer_large_patch4_window7_224"],
use_ssld=use_ssld)
return model
def SwinTransformer_large_patch4_window12_384(pretrained=False, use_ssld=False, **kwargs):
def SwinTransformer_large_patch4_window12_384(pretrained=False,
use_ssld=False,
**kwargs):
model = SwinTransformer(
img_size=384,
embed_dim=192,
@ -810,5 +848,9 @@ def SwinTransformer_large_patch4_window12_384(pretrained=False, use_ssld=False,
num_heads=[6, 12, 24, 48],
window_size=12,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["SwinTransformer_large_patch4_window12_384"], use_ssld=use_ssld)
_load_pretrained(
pretrained,
model,
MODEL_URLS["SwinTransformer_large_patch4_window12_384"],
use_ssld=use_ssld)
return model

216
ppcls/arch/backbone/model_zoo/tnt.py
View File

@ -23,16 +23,13 @@ from paddle.nn.initializer import TruncatedNormal, Constant
from ppcls.arch.backbone.base.theseus_layer import Identity
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"TNT_small":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/TNT_small_pretrained.pdparams"
}
__all__ = MODEL_URLS.keys()
trunc_normal_ = TruncatedNormal(std=.02)
zeros_ = Constant(value=0.)
ones_ = Constant(value=1.)
@ -66,8 +63,12 @@ class DropPath(nn.Layer):
class Mlp(nn.Layer):
def __init__(self, in_features, hidden_features=None,
out_features=None, act_layer=nn.GELU, drop=0.):
def __init__(self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
drop=0.):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
@ -86,14 +87,19 @@ class Mlp(nn.Layer):
class Attention(nn.Layer):
def __init__(self, dim, hidden_dim, num_heads=8,
qkv_bias=False, attn_drop=0., proj_drop=0.):
def __init__(self,
dim,
hidden_dim,
num_heads=8,
qkv_bias=False,
attn_drop=0.,
proj_drop=0.):
super().__init__()
self.hidden_dim = hidden_dim
self.num_heads = num_heads
head_dim = hidden_dim // num_heads
self.head_dim = head_dim
self.scale = head_dim ** -0.5
self.scale = head_dim**-0.5
self.qk = nn.Linear(dim, hidden_dim * 2, bias_attr=qkv_bias)
self.v = nn.Linear(dim, dim, bias_attr=qkv_bias)
@ -103,73 +109,103 @@ class Attention(nn.Layer):
def forward(self, x):
B, N, C = x.shape
qk = self.qk(x).reshape((B, N, 2, self.num_heads, self.head_dim)).transpose((2, 0, 3, 1, 4))
qk = self.qk(x).reshape(
(B, N, 2, self.num_heads, self.head_dim)).transpose(
(2, 0, 3, 1, 4))
q, k = qk[0], qk[1]
v = self.v(x).reshape((B, N, self.num_heads, -1)).transpose((0, 2, 1, 3))
v = self.v(x).reshape((B, N, self.num_heads, -1)).transpose(
(0, 2, 1, 3))
attn = (q @ k.transpose((0, 1, 3, 2))) * self.scale
attn = (q @k.transpose((0, 1, 3, 2))) * self.scale
attn = nn.functional.softmax(attn, axis=-1)
attn = self.attn_drop(attn)
x = (attn @ v).transpose((0, 2, 1, 3)).reshape((B, N, -1))
x = (attn @v).transpose((0, 2, 1, 3)).reshape((B, N, -1))
x = self.proj(x)
x = self.proj_drop(x)
return x
class Block(nn.Layer):
def __init__(self, dim, in_dim, num_pixel, num_heads=12, in_num_head=4, mlp_ratio=4.,
qkv_bias=False, drop=0., attn_drop=0., drop_path=0., act_layer=nn.GELU,
def __init__(self,
dim,
in_dim,
num_pixel,
num_heads=12,
in_num_head=4,
mlp_ratio=4.,
qkv_bias=False,
drop=0.,
attn_drop=0.,
drop_path=0.,
act_layer=nn.GELU,
norm_layer=nn.LayerNorm):
super().__init__()
# Inner transformer
self.norm_in = norm_layer(in_dim)
self.attn_in = Attention(
in_dim, in_dim, num_heads=in_num_head,
qkv_bias=qkv_bias, attn_drop=attn_drop,
proj_drop=drop
)
in_dim,
in_dim,
num_heads=in_num_head,
qkv_bias=qkv_bias,
attn_drop=attn_drop,
proj_drop=drop)
self.norm_mlp_in = norm_layer(in_dim)
self.mlp_in = Mlp(
in_features=in_dim, hidden_features=int(in_dim * 4),
out_features=in_dim, act_layer=act_layer, drop=drop
)
self.mlp_in = Mlp(in_features=in_dim,
hidden_features=int(in_dim * 4),
out_features=in_dim,
act_layer=act_layer,
drop=drop)
self.norm1_proj = norm_layer(in_dim)
self.proj = nn.Linear(in_dim * num_pixel, dim)
# Outer transformer
self.norm_out = norm_layer(dim)
self.attn_out = Attention(
dim, dim, num_heads=num_heads, qkv_bias=qkv_bias,
attn_drop=attn_drop, proj_drop=drop
)
dim,
dim,
num_heads=num_heads,
qkv_bias=qkv_bias,
attn_drop=attn_drop,
proj_drop=drop)
self.drop_path = DropPath(drop_path) if drop_path > 0. else Identity()
self.norm_mlp = norm_layer(dim)
self.mlp = Mlp(
in_features=dim, hidden_features=int(dim * mlp_ratio),
out_features=dim, act_layer=act_layer, drop=drop
)
self.mlp = Mlp(in_features=dim,
hidden_features=int(dim * mlp_ratio),
out_features=dim,
act_layer=act_layer,
drop=drop)
def forward(self, pixel_embed, patch_embed):
# inner
pixel_embed = pixel_embed + self.drop_path(self.attn_in(self.norm_in(pixel_embed)))
pixel_embed = pixel_embed + self.drop_path(self.mlp_in(self.norm_mlp_in(pixel_embed)))
pixel_embed = pixel_embed + self.drop_path(
self.attn_in(self.norm_in(pixel_embed)))
pixel_embed = pixel_embed + self.drop_path(
self.mlp_in(self.norm_mlp_in(pixel_embed)))
# outer
B, N, C = patch_embed.shape
patch_embed[:, 1:] = patch_embed[:, 1:] + self.proj(self.norm1_proj(pixel_embed).reshape((B, N - 1, -1)))
patch_embed = patch_embed + self.drop_path(self.attn_out(self.norm_out(patch_embed)))
patch_embed = patch_embed + self.drop_path(self.mlp(self.norm_mlp(patch_embed)))
patch_embed[:, 1:] = patch_embed[:, 1:] + self.proj(
self.norm1_proj(pixel_embed).reshape((B, N - 1, -1)))
patch_embed = patch_embed + self.drop_path(
self.attn_out(self.norm_out(patch_embed)))
patch_embed = patch_embed + self.drop_path(
self.mlp(self.norm_mlp(patch_embed)))
return pixel_embed, patch_embed
class PixelEmbed(nn.Layer):
def __init__(self, img_size=224, patch_size=16, in_chans=3, in_dim=48, stride=4):
def __init__(self,
img_size=224,
patch_size=16,
in_chans=3,
in_dim=48,
stride=4):
super().__init__()
num_patches = (img_size // patch_size) ** 2
num_patches = (img_size // patch_size)**2
self.img_size = img_size
self.num_patches = num_patches
self.in_dim = in_dim
@ -177,10 +213,7 @@ class PixelEmbed(nn.Layer):
self.new_patch_size = new_patch_size
self.proj = nn.Conv2D(
in_chans, self.in_dim,
kernel_size=7, padding=3,
stride=stride
)
in_chans, self.in_dim, kernel_size=7, padding=3, stride=stride)
def forward(self, x, pixel_pos):
B, C, H, W = x.shape
@ -188,50 +221,64 @@ class PixelEmbed(nn.Layer):
x = self.proj(x)
x = nn.functional.unfold(x, self.new_patch_size, self.new_patch_size)
x = x.transpose((0, 2, 1)).reshape((B * self.num_patches, self.in_dim, self.new_patch_size, self.new_patch_size))
x = x.transpose((0, 2, 1)).reshape(
(B * self.num_patches, self.in_dim, self.new_patch_size,
self.new_patch_size))
x = x + pixel_pos
x = x.reshape((B * self.num_patches, self.in_dim, -1)).transpose((0, 2, 1))
x = x.reshape((B * self.num_patches, self.in_dim, -1)).transpose(
(0, 2, 1))
return x
class TNT(nn.Layer):
def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, in_dim=48, depth=12,
num_heads=12, in_num_head=4, mlp_ratio=4., qkv_bias=False, drop_rate=0., attn_drop_rate=0.,
drop_path_rate=0., norm_layer=nn.LayerNorm, first_stride=4, class_dim=1000):
def __init__(self,
img_size=224,
patch_size=16,
in_chans=3,
embed_dim=768,
in_dim=48,
depth=12,
num_heads=12,
in_num_head=4,
mlp_ratio=4.,
qkv_bias=False,
drop_rate=0.,
attn_drop_rate=0.,
drop_path_rate=0.,
norm_layer=nn.LayerNorm,
first_stride=4,
class_num=1000):
super().__init__()
self.class_dim = class_dim
self.class_num = class_num
# num_features for consistency with other models
self.num_features = self.embed_dim = embed_dim
self.pixel_embed = PixelEmbed(
img_size=img_size, patch_size=patch_size,
in_chans=in_chans, in_dim=in_dim, stride=first_stride
)
img_size=img_size,
patch_size=patch_size,
in_chans=in_chans,
in_dim=in_dim,
stride=first_stride)
num_patches = self.pixel_embed.num_patches
self.num_patches = num_patches
new_patch_size = self.pixel_embed.new_patch_size
num_pixel = new_patch_size ** 2
num_pixel = new_patch_size**2
self.norm1_proj = norm_layer(num_pixel * in_dim)
self.proj = nn.Linear(num_pixel * in_dim, embed_dim)
self.norm2_proj = norm_layer(embed_dim)
self.cls_token = self.create_parameter(
shape=(1, 1, embed_dim),
default_initializer=zeros_
)
shape=(1, 1, embed_dim), default_initializer=zeros_)
self.add_parameter("cls_token", self.cls_token)
self.patch_pos = self.create_parameter(
shape=(1, num_patches + 1, embed_dim),
default_initializer=zeros_
)
shape=(1, num_patches + 1, embed_dim), default_initializer=zeros_)
self.add_parameter("patch_pos", self.patch_pos)
self.pixel_pos = self.create_parameter(
shape=(1, in_dim, new_patch_size, new_patch_size),
default_initializer=zeros_
)
shape=(1, in_dim, new_patch_size, new_patch_size),
default_initializer=zeros_)
self.add_parameter("pixel_pos", self.pixel_pos)
self.pos_drop = nn.Dropout(p=drop_rate)
@ -241,17 +288,24 @@ class TNT(nn.Layer):
blocks = []
for i in range(depth):
blocks.append(Block(
dim=embed_dim, in_dim=in_dim, num_pixel=num_pixel, num_heads=num_heads,
in_num_head=in_num_head, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias,
drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i],
norm_layer=norm_layer
))
blocks.append(
Block(
dim=embed_dim,
in_dim=in_dim,
num_pixel=num_pixel,
num_heads=num_heads,
in_num_head=in_num_head,
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
drop=drop_rate,
attn_drop=attn_drop_rate,
drop_path=dpr[i],
norm_layer=norm_layer))
self.blocks = nn.LayerList(blocks)
self.norm = norm_layer(embed_dim)
if class_dim > 0:
self.head = nn.Linear(embed_dim, class_dim)
if class_num > 0:
self.head = nn.Linear(embed_dim, class_num)
trunc_normal_(self.cls_token)
trunc_normal_(self.patch_pos)
@ -271,8 +325,12 @@ class TNT(nn.Layer):
B = x.shape[0]
pixel_embed = self.pixel_embed(x, self.pixel_pos)
patch_embed = self.norm2_proj(self.proj(self.norm1_proj(pixel_embed.reshape((B, self.num_patches, -1)))))
patch_embed = paddle.concat((self.cls_token.expand((B, -1, -1)), patch_embed), axis=1)
patch_embed = self.norm2_proj(
self.proj(
self.norm1_proj(
pixel_embed.reshape((B, self.num_patches, -1)))))
patch_embed = paddle.concat(
(self.cls_token.expand((B, -1, -1)), patch_embed), axis=1)
patch_embed = patch_embed + self.patch_pos
patch_embed = self.pos_drop(patch_embed)
@ -285,7 +343,7 @@ class TNT(nn.Layer):
def forward(self, x):
x = self.forward_features(x)
if self.class_dim > 0:
if self.class_num > 0:
x = self.head(x)
return x
@ -304,15 +362,13 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def TNT_small(pretrained=False, **kwargs):
model = TNT(
patch_size=16,
embed_dim=384,
in_dim=24,
depth=12,
num_heads=6,
in_num_head=4,
qkv_bias=False,
**kwargs
)
model = TNT(patch_size=16,
embed_dim=384,
in_dim=24,
depth=12,
num_heads=6,
in_num_head=4,
qkv_bias=False,
**kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["TNT_small"])
return model

6
ppcls/arch/backbone/model_zoo/vision_transformer.py
View File

@ -231,7 +231,7 @@ class VisionTransformer(nn.Layer):
img_size=224,
patch_size=16,
in_chans=3,
class_dim=1000,
class_num=1000,
embed_dim=768,
depth=12,
num_heads=12,
@ -245,7 +245,7 @@ class VisionTransformer(nn.Layer):
epsilon=1e-5,
**args):
super().__init__()
self.class_dim = class_dim
self.class_num = class_num
self.num_features = self.embed_dim = embed_dim
@ -284,7 +284,7 @@ class VisionTransformer(nn.Layer):
# Classifier head
self.head = nn.Linear(embed_dim,
class_dim) if class_dim > 0 else Identity()
class_num) if class_num > 0 else Identity()
trunc_normal_(self.pos_embed)
trunc_normal_(self.cls_token)

38
ppcls/arch/backbone/model_zoo/xception.py
View File

@ -8,14 +8,16 @@ from paddle.nn.initializer import Uniform
import math
import sys
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {
"Xception41": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception41_pretrained.pdparams",
"Xception65": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception65_pretrained.pdparams",
"Xception71": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception71_pretrained.pdparams"
}
"Xception41":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception41_pretrained.pdparams",
"Xception65":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception65_pretrained.pdparams",
"Xception71":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception71_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
@ -290,7 +292,7 @@ class ExitFlowBottleneckBlock(nn.Layer):
class ExitFlow(nn.Layer):
def __init__(self, class_dim):
def __init__(self, class_num):
super(ExitFlow, self).__init__()
name = "exit_flow"
@ -303,7 +305,7 @@ class ExitFlow(nn.Layer):
stdv = 1.0 / math.sqrt(2048 * 1.0)
self._out = Linear(
2048,
class_dim,
class_num,
weight_attr=ParamAttr(
name="fc_weights", initializer=Uniform(-stdv, stdv)),
bias_attr=ParamAttr(name="fc_offset"))
@ -324,13 +326,13 @@ class Xception(nn.Layer):
def __init__(self,
entry_flow_block_num=3,
middle_flow_block_num=8,
class_dim=1000):
class_num=1000):
super(Xception, self).__init__()
self.entry_flow_block_num = entry_flow_block_num
self.middle_flow_block_num = middle_flow_block_num
self._entry_flow = EntryFlow(entry_flow_block_num)
self._middle_flow = MiddleFlow(middle_flow_block_num)
self._exit_flow = ExitFlow(class_dim)
self._exit_flow = ExitFlow(class_num)
def forward(self, inputs):
x = self._entry_flow(inputs)
@ -338,6 +340,7 @@ class Xception(nn.Layer):
x = self._exit_flow(x)
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -349,21 +352,26 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
raise RuntimeError(
"pretrained type is not available. Please use `string` or `boolean` type."
)
def Xception41(pretrained=False, use_ssld=False, **kwargs):
model = Xception(entry_flow_block_num=3, middle_flow_block_num=8, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Xception41"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["Xception41"], use_ssld=use_ssld)
return model
def Xception65(pretrained=False, use_ssld=False, **kwargs):
model = Xception(entry_flow_block_num=3, middle_flow_block_num=16, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Xception65"], use_ssld=use_ssld)
model = Xception(
entry_flow_block_num=3, middle_flow_block_num=16, **kwargs)
_load_pretrained(
pretrained, model, MODEL_URLS["Xception65"], use_ssld=use_ssld)
return model
def Xception71(pretrained=False, use_ssld=False, **kwargs):
model = Xception(entry_flow_block_num=5, middle_flow_block_num=16, **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Xception71"], use_ssld=use_ssld)
model = Xception(
entry_flow_block_num=5, middle_flow_block_num=16, **kwargs)
_load_pretrained(
pretrained, model, MODEL_URLS["Xception71"], use_ssld=use_ssld)
return model

22
ppcls/arch/backbone/model_zoo/xception_deeplab.py
View File

@ -21,8 +21,12 @@ from paddle.nn import AdaptiveAvgPool2D, MaxPool2D, AvgPool2D
from ppcls.utils.save_load import load_dygraph_pretrain, load_dygraph_pretrain_from_url
MODEL_URLS = {"Xception41_deeplab": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception41_deeplab_pretrained.pdparams",
"Xception65_deeplab": "https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception65_deeplab_pretrained.pdparams"}
MODEL_URLS = {
"Xception41_deeplab":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception41_deeplab_pretrained.pdparams",
"Xception65_deeplab":
"https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/Xception65_deeplab_pretrained.pdparams"
}
__all__ = list(MODEL_URLS.keys())
@ -268,7 +272,7 @@ class Xception_Block(nn.Layer):
class XceptionDeeplab(nn.Layer):
def __init__(self, backbone, class_dim=1000):
def __init__(self, backbone, class_num=1000):
super(XceptionDeeplab, self).__init__()
bottleneck_params = gen_bottleneck_params(backbone)
@ -370,7 +374,7 @@ class XceptionDeeplab(nn.Layer):
self._pool = AdaptiveAvgPool2D(1)
self._fc = Linear(
self.chns[1][-1],
class_dim,
class_num,
weight_attr=ParamAttr(name="fc_weights"),
bias_attr=ParamAttr(name="fc_bias"))
@ -388,8 +392,8 @@ class XceptionDeeplab(nn.Layer):
x = paddle.squeeze(x, axis=[2, 3])
x = self._fc(x)
return x
def _load_pretrained(pretrained, model, model_url, use_ssld=False):
if pretrained is False:
pass
@ -405,11 +409,13 @@ def _load_pretrained(pretrained, model, model_url, use_ssld=False):
def Xception41_deeplab(pretrained=False, use_ssld=False, **kwargs):
model = XceptionDeeplab('xception_41', **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Xception41_deeplab"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["Xception41_deeplab"], use_ssld=use_ssld)
return model
def Xception65_deeplab(pretrained=False, use_ssld=False, **kwargs):
model = XceptionDeeplab("xception_65", **kwargs)
_load_pretrained(pretrained, model, MODEL_URLS["Xception65_deeplab"], use_ssld=use_ssld)
_load_pretrained(
pretrained, model, MODEL_URLS["Xception65_deeplab"], use_ssld=use_ssld)
return model