2020-04-24 12:14:56 +08:00
|
|
|
# encoding: utf-8
|
|
|
|
"""
|
|
|
|
@author: xingyu liao
|
2020-07-29 17:43:39 +08:00
|
|
|
@contact: sherlockliao01@gmail.com
|
2020-04-24 12:14:56 +08:00
|
|
|
"""
|
|
|
|
|
|
|
|
# based on:
|
|
|
|
# https://github.com/XingangPan/IBN-Net/blob/master/models/imagenet/resnext_ibn_a.py
|
|
|
|
|
|
|
|
import logging
|
2020-07-29 17:43:39 +08:00
|
|
|
import math
|
|
|
|
|
2020-04-24 12:14:56 +08:00
|
|
|
import torch
|
2020-07-29 17:43:39 +08:00
|
|
|
import torch.nn as nn
|
|
|
|
|
2020-09-01 16:14:45 +08:00
|
|
|
from fastreid.layers import *
|
2020-07-31 10:42:38 +08:00
|
|
|
from fastreid.utils import comm
|
2020-09-01 16:14:45 +08:00
|
|
|
from fastreid.utils.checkpoint import get_missing_parameters_message, get_unexpected_parameters_message
|
2020-04-24 12:14:56 +08:00
|
|
|
from .build import BACKBONE_REGISTRY
|
|
|
|
|
2020-07-29 17:43:39 +08:00
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
model_urls = {
|
|
|
|
'ibn_101x': 'https://github.com/XingangPan/IBN-Net/releases/download/v1.0/resnext101_ibn_a-6ace051d.pth',
|
|
|
|
}
|
|
|
|
|
2020-04-24 12:14:56 +08:00
|
|
|
|
|
|
|
class Bottleneck(nn.Module):
|
|
|
|
"""
|
|
|
|
RexNeXt bottleneck type C
|
|
|
|
"""
|
|
|
|
expansion = 4
|
|
|
|
|
2020-07-29 17:43:39 +08:00
|
|
|
def __init__(self, inplanes, planes, bn_norm, num_splits, with_ibn, baseWidth, cardinality, stride=1,
|
|
|
|
downsample=None):
|
2020-04-24 12:14:56 +08:00
|
|
|
""" Constructor
|
|
|
|
Args:
|
|
|
|
inplanes: input channel dimensionality
|
|
|
|
planes: output channel dimensionality
|
|
|
|
baseWidth: base width.
|
|
|
|
cardinality: num of convolution groups.
|
|
|
|
stride: conv stride. Replaces pooling layer.
|
|
|
|
"""
|
|
|
|
super(Bottleneck, self).__init__()
|
|
|
|
|
|
|
|
D = int(math.floor(planes * (baseWidth / 64)))
|
|
|
|
C = cardinality
|
|
|
|
self.conv1 = nn.Conv2d(inplanes, D * C, kernel_size=1, stride=1, padding=0, bias=False)
|
|
|
|
if with_ibn:
|
2020-07-29 17:43:39 +08:00
|
|
|
self.bn1 = IBN(D * C, bn_norm, num_splits)
|
2020-04-24 12:14:56 +08:00
|
|
|
else:
|
2020-07-29 17:43:39 +08:00
|
|
|
self.bn1 = get_norm(bn_norm, D * C, num_splits)
|
2020-04-24 12:14:56 +08:00
|
|
|
self.conv2 = nn.Conv2d(D * C, D * C, kernel_size=3, stride=stride, padding=1, groups=C, bias=False)
|
2020-07-29 17:43:39 +08:00
|
|
|
self.bn2 = get_norm(bn_norm, D * C, num_splits)
|
2020-04-24 12:14:56 +08:00
|
|
|
self.conv3 = nn.Conv2d(D * C, planes * 4, kernel_size=1, stride=1, padding=0, bias=False)
|
2020-07-29 17:43:39 +08:00
|
|
|
self.bn3 = get_norm(bn_norm, planes * 4, num_splits)
|
2020-04-24 12:14:56 +08:00
|
|
|
self.relu = nn.ReLU(inplace=True)
|
|
|
|
|
|
|
|
self.downsample = downsample
|
|
|
|
|
|
|
|
def forward(self, x):
|
|
|
|
residual = x
|
|
|
|
|
|
|
|
out = self.conv1(x)
|
|
|
|
out = self.bn1(out)
|
|
|
|
out = self.relu(out)
|
|
|
|
|
|
|
|
out = self.conv2(out)
|
|
|
|
out = self.bn2(out)
|
|
|
|
out = self.relu(out)
|
|
|
|
|
|
|
|
out = self.conv3(out)
|
|
|
|
out = self.bn3(out)
|
|
|
|
|
|
|
|
if self.downsample is not None:
|
|
|
|
residual = self.downsample(x)
|
|
|
|
|
|
|
|
out += residual
|
|
|
|
out = self.relu(out)
|
|
|
|
|
|
|
|
return out
|
|
|
|
|
|
|
|
|
|
|
|
class ResNeXt(nn.Module):
|
|
|
|
"""
|
|
|
|
ResNext optimized for the ImageNet dataset, as specified in
|
|
|
|
https://arxiv.org/pdf/1611.05431.pdf
|
|
|
|
"""
|
|
|
|
|
2020-09-01 16:14:45 +08:00
|
|
|
def __init__(self, last_stride, bn_norm, num_splits, with_ibn, with_nl, block, layers, non_layers,
|
|
|
|
baseWidth=4, cardinality=32):
|
2020-04-24 12:14:56 +08:00
|
|
|
""" Constructor
|
|
|
|
Args:
|
|
|
|
baseWidth: baseWidth for ResNeXt.
|
|
|
|
cardinality: number of convolution groups.
|
|
|
|
layers: config of layers, e.g., [3, 4, 6, 3]
|
|
|
|
"""
|
|
|
|
super(ResNeXt, self).__init__()
|
|
|
|
|
|
|
|
self.cardinality = cardinality
|
|
|
|
self.baseWidth = baseWidth
|
|
|
|
self.inplanes = 64
|
|
|
|
self.output_size = 64
|
|
|
|
|
|
|
|
self.conv1 = nn.Conv2d(3, 64, 7, 2, 3, bias=False)
|
2020-07-29 17:43:39 +08:00
|
|
|
self.bn1 = get_norm(bn_norm, 64, num_splits)
|
2020-04-24 12:14:56 +08:00
|
|
|
self.relu = nn.ReLU(inplace=True)
|
|
|
|
self.maxpool1 = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
|
2020-07-29 17:43:39 +08:00
|
|
|
self.layer1 = self._make_layer(block, 64, layers[0], 1, bn_norm, num_splits, with_ibn=with_ibn)
|
|
|
|
self.layer2 = self._make_layer(block, 128, layers[1], 2, bn_norm, num_splits, with_ibn=with_ibn)
|
|
|
|
self.layer3 = self._make_layer(block, 256, layers[2], 2, bn_norm, num_splits, with_ibn=with_ibn)
|
|
|
|
self.layer4 = self._make_layer(block, 512, layers[3], last_stride, bn_norm, num_splits, with_ibn=with_ibn)
|
2020-04-24 12:14:56 +08:00
|
|
|
|
|
|
|
self.random_init()
|
|
|
|
|
2020-09-01 16:14:45 +08:00
|
|
|
# fmt: off
|
|
|
|
if with_nl: self._build_nonlocal(layers, non_layers, bn_norm, num_splits)
|
|
|
|
else: self.NL_1_idx = self.NL_2_idx = self.NL_3_idx = self.NL_4_idx = []
|
|
|
|
# fmt: on
|
|
|
|
|
2020-07-29 17:43:39 +08:00
|
|
|
def _make_layer(self, block, planes, blocks, stride=1, bn_norm='BN', num_splits=1, with_ibn=False):
|
2020-04-24 12:14:56 +08:00
|
|
|
""" Stack n bottleneck modules where n is inferred from the depth of the network.
|
|
|
|
Args:
|
|
|
|
block: block type used to construct ResNext
|
|
|
|
planes: number of output channels (need to multiply by block.expansion)
|
|
|
|
blocks: number of blocks to be built
|
|
|
|
stride: factor to reduce the spatial dimensionality in the first bottleneck of the block.
|
|
|
|
Returns: a Module consisting of n sequential bottlenecks.
|
|
|
|
"""
|
|
|
|
downsample = None
|
|
|
|
if stride != 1 or self.inplanes != planes * block.expansion:
|
|
|
|
downsample = nn.Sequential(
|
|
|
|
nn.Conv2d(self.inplanes, planes * block.expansion,
|
|
|
|
kernel_size=1, stride=stride, bias=False),
|
2020-07-29 17:43:39 +08:00
|
|
|
get_norm(bn_norm, planes * block.expansion, num_splits),
|
2020-04-24 12:14:56 +08:00
|
|
|
)
|
|
|
|
|
|
|
|
layers = []
|
|
|
|
if planes == 512:
|
|
|
|
with_ibn = False
|
2020-07-29 17:43:39 +08:00
|
|
|
layers.append(block(self.inplanes, planes, bn_norm, num_splits, with_ibn,
|
|
|
|
self.baseWidth, self.cardinality, stride, downsample))
|
2020-04-24 12:14:56 +08:00
|
|
|
self.inplanes = planes * block.expansion
|
|
|
|
for i in range(1, blocks):
|
2020-07-29 17:43:39 +08:00
|
|
|
layers.append(
|
|
|
|
block(self.inplanes, planes, bn_norm, num_splits, with_ibn, self.baseWidth, self.cardinality, 1, None))
|
2020-04-24 12:14:56 +08:00
|
|
|
|
|
|
|
return nn.Sequential(*layers)
|
|
|
|
|
2020-09-01 16:14:45 +08:00
|
|
|
def _build_nonlocal(self, layers, non_layers, bn_norm, num_splits):
|
|
|
|
self.NL_1 = nn.ModuleList(
|
|
|
|
[Non_local(256, bn_norm, num_splits) for _ in range(non_layers[0])])
|
|
|
|
self.NL_1_idx = sorted([layers[0] - (i + 1) for i in range(non_layers[0])])
|
|
|
|
self.NL_2 = nn.ModuleList(
|
|
|
|
[Non_local(512, bn_norm, num_splits) for _ in range(non_layers[1])])
|
|
|
|
self.NL_2_idx = sorted([layers[1] - (i + 1) for i in range(non_layers[1])])
|
|
|
|
self.NL_3 = nn.ModuleList(
|
|
|
|
[Non_local(1024, bn_norm, num_splits) for _ in range(non_layers[2])])
|
|
|
|
self.NL_3_idx = sorted([layers[2] - (i + 1) for i in range(non_layers[2])])
|
|
|
|
self.NL_4 = nn.ModuleList(
|
|
|
|
[Non_local(2048, bn_norm, num_splits) for _ in range(non_layers[3])])
|
|
|
|
self.NL_4_idx = sorted([layers[3] - (i + 1) for i in range(non_layers[3])])
|
|
|
|
|
2020-04-24 12:14:56 +08:00
|
|
|
def forward(self, x):
|
|
|
|
x = self.conv1(x)
|
|
|
|
x = self.bn1(x)
|
|
|
|
x = self.relu(x)
|
|
|
|
x = self.maxpool1(x)
|
|
|
|
|
2020-09-01 16:14:45 +08:00
|
|
|
NL1_counter = 0
|
|
|
|
if len(self.NL_1_idx) == 0:
|
|
|
|
self.NL_1_idx = [-1]
|
|
|
|
for i in range(len(self.layer1)):
|
|
|
|
x = self.layer1[i](x)
|
|
|
|
if i == self.NL_1_idx[NL1_counter]:
|
|
|
|
_, C, H, W = x.shape
|
|
|
|
x = self.NL_1[NL1_counter](x)
|
|
|
|
NL1_counter += 1
|
|
|
|
# Layer 2
|
|
|
|
NL2_counter = 0
|
|
|
|
if len(self.NL_2_idx) == 0:
|
|
|
|
self.NL_2_idx = [-1]
|
|
|
|
for i in range(len(self.layer2)):
|
|
|
|
x = self.layer2[i](x)
|
|
|
|
if i == self.NL_2_idx[NL2_counter]:
|
|
|
|
_, C, H, W = x.shape
|
|
|
|
x = self.NL_2[NL2_counter](x)
|
|
|
|
NL2_counter += 1
|
|
|
|
# Layer 3
|
|
|
|
NL3_counter = 0
|
|
|
|
if len(self.NL_3_idx) == 0:
|
|
|
|
self.NL_3_idx = [-1]
|
|
|
|
for i in range(len(self.layer3)):
|
|
|
|
x = self.layer3[i](x)
|
|
|
|
if i == self.NL_3_idx[NL3_counter]:
|
|
|
|
_, C, H, W = x.shape
|
|
|
|
x = self.NL_3[NL3_counter](x)
|
|
|
|
NL3_counter += 1
|
|
|
|
# Layer 4
|
|
|
|
NL4_counter = 0
|
|
|
|
if len(self.NL_4_idx) == 0:
|
|
|
|
self.NL_4_idx = [-1]
|
|
|
|
for i in range(len(self.layer4)):
|
|
|
|
x = self.layer4[i](x)
|
|
|
|
if i == self.NL_4_idx[NL4_counter]:
|
|
|
|
_, C, H, W = x.shape
|
|
|
|
x = self.NL_4[NL4_counter](x)
|
|
|
|
NL4_counter += 1
|
2020-04-24 12:14:56 +08:00
|
|
|
return x
|
|
|
|
|
|
|
|
def random_init(self):
|
|
|
|
self.conv1.weight.data.normal_(0, math.sqrt(2. / (7 * 7 * 64)))
|
|
|
|
for m in self.modules():
|
|
|
|
if isinstance(m, nn.Conv2d):
|
|
|
|
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
|
|
|
|
m.weight.data.normal_(0, math.sqrt(2. / n))
|
|
|
|
elif isinstance(m, nn.BatchNorm2d):
|
|
|
|
m.weight.data.fill_(1)
|
|
|
|
m.bias.data.zero_()
|
|
|
|
elif isinstance(m, nn.InstanceNorm2d):
|
|
|
|
m.weight.data.fill_(1)
|
|
|
|
m.bias.data.zero_()
|
|
|
|
|
|
|
|
|
2020-07-29 17:43:39 +08:00
|
|
|
def init_pretrained_weights(key):
|
|
|
|
"""Initializes model with pretrained weights.
|
|
|
|
|
|
|
|
Layers that don't match with pretrained layers in name or size are kept unchanged.
|
|
|
|
"""
|
|
|
|
import os
|
|
|
|
import errno
|
|
|
|
import gdown
|
|
|
|
|
|
|
|
def _get_torch_home():
|
|
|
|
ENV_TORCH_HOME = 'TORCH_HOME'
|
|
|
|
ENV_XDG_CACHE_HOME = 'XDG_CACHE_HOME'
|
|
|
|
DEFAULT_CACHE_DIR = '~/.cache'
|
|
|
|
torch_home = os.path.expanduser(
|
|
|
|
os.getenv(
|
|
|
|
ENV_TORCH_HOME,
|
|
|
|
os.path.join(
|
|
|
|
os.getenv(ENV_XDG_CACHE_HOME, DEFAULT_CACHE_DIR), 'torch'
|
|
|
|
)
|
|
|
|
)
|
|
|
|
)
|
|
|
|
return torch_home
|
|
|
|
|
|
|
|
torch_home = _get_torch_home()
|
|
|
|
model_dir = os.path.join(torch_home, 'checkpoints')
|
|
|
|
try:
|
|
|
|
os.makedirs(model_dir)
|
|
|
|
except OSError as e:
|
|
|
|
if e.errno == errno.EEXIST:
|
|
|
|
# Directory already exists, ignore.
|
|
|
|
pass
|
|
|
|
else:
|
|
|
|
# Unexpected OSError, re-raise.
|
|
|
|
raise
|
|
|
|
|
|
|
|
filename = model_urls[key].split('/')[-1]
|
|
|
|
|
|
|
|
cached_file = os.path.join(model_dir, filename)
|
|
|
|
|
|
|
|
if not os.path.exists(cached_file):
|
2020-07-31 10:42:38 +08:00
|
|
|
if comm.is_main_process():
|
|
|
|
gdown.download(model_urls[key], cached_file, quiet=False)
|
2020-08-04 15:56:36 +08:00
|
|
|
|
|
|
|
comm.synchronize()
|
2020-07-29 17:43:39 +08:00
|
|
|
|
|
|
|
logger.info(f"Loading pretrained model from {cached_file}")
|
2020-07-31 10:42:38 +08:00
|
|
|
state_dict = torch.load(cached_file, map_location=torch.device('cpu'))
|
2020-07-29 17:43:39 +08:00
|
|
|
|
|
|
|
return state_dict
|
|
|
|
|
2020-08-04 15:56:36 +08:00
|
|
|
|
2020-04-24 12:14:56 +08:00
|
|
|
@BACKBONE_REGISTRY.register()
|
|
|
|
def build_resnext_backbone(cfg):
|
|
|
|
"""
|
|
|
|
Create a ResNeXt instance from config.
|
|
|
|
Returns:
|
|
|
|
ResNeXt: a :class:`ResNeXt` instance.
|
|
|
|
"""
|
|
|
|
|
|
|
|
# fmt: off
|
2020-09-01 16:14:45 +08:00
|
|
|
pretrain = cfg.MODEL.BACKBONE.PRETRAIN
|
2020-04-24 12:14:56 +08:00
|
|
|
pretrain_path = cfg.MODEL.BACKBONE.PRETRAIN_PATH
|
2020-09-01 16:14:45 +08:00
|
|
|
last_stride = cfg.MODEL.BACKBONE.LAST_STRIDE
|
|
|
|
bn_norm = cfg.MODEL.BACKBONE.NORM
|
|
|
|
num_splits = cfg.MODEL.BACKBONE.NORM_SPLIT
|
|
|
|
with_ibn = cfg.MODEL.BACKBONE.WITH_IBN
|
|
|
|
with_nl = cfg.MODEL.BACKBONE.WITH_NL
|
|
|
|
depth = cfg.MODEL.BACKBONE.DEPTH
|
|
|
|
# fmt: on
|
|
|
|
|
|
|
|
num_blocks_per_stage = {
|
|
|
|
'50x': [3, 4, 6, 3],
|
|
|
|
'101x': [3, 4, 23, 3],
|
|
|
|
'152x': [3, 8, 36, 3], }[depth]
|
|
|
|
nl_layers_per_stage = {
|
|
|
|
'50x': [0, 2, 3, 0],
|
|
|
|
'101x': [0, 2, 3, 0]}[depth]
|
|
|
|
model = ResNeXt(last_stride, bn_norm, num_splits, with_ibn, with_nl, Bottleneck,
|
|
|
|
num_blocks_per_stage, nl_layers_per_stage)
|
2020-04-24 12:14:56 +08:00
|
|
|
if pretrain:
|
2020-07-29 17:43:39 +08:00
|
|
|
if pretrain_path:
|
|
|
|
try:
|
|
|
|
state_dict = torch.load(pretrain_path, map_location=torch.device('cpu'))['model']
|
|
|
|
# Remove module.encoder in name
|
|
|
|
new_state_dict = {}
|
|
|
|
for k in state_dict:
|
|
|
|
new_k = '.'.join(k.split('.')[2:])
|
|
|
|
if new_k in model.state_dict() and (model.state_dict()[new_k].shape == state_dict[k].shape):
|
|
|
|
new_state_dict[new_k] = state_dict[k]
|
|
|
|
state_dict = new_state_dict
|
|
|
|
logger.info(f"Loading pretrained model from {pretrain_path}")
|
|
|
|
except FileNotFoundError as e:
|
|
|
|
logger.info(f'{pretrain_path} is not found! Please check this path.')
|
|
|
|
raise e
|
|
|
|
except KeyError as e:
|
|
|
|
logger.info("State dict keys error! Please check the state dict.")
|
|
|
|
raise e
|
|
|
|
else:
|
|
|
|
key = depth
|
|
|
|
if with_ibn: key = 'ibn_' + key
|
|
|
|
|
|
|
|
state_dict = init_pretrained_weights(key)
|
|
|
|
|
|
|
|
incompatible = model.load_state_dict(state_dict, strict=False)
|
|
|
|
if incompatible.missing_keys:
|
|
|
|
logger.info(
|
|
|
|
get_missing_parameters_message(incompatible.missing_keys)
|
|
|
|
)
|
|
|
|
if incompatible.unexpected_keys:
|
|
|
|
logger.info(
|
|
|
|
get_unexpected_parameters_message(incompatible.unexpected_keys)
|
|
|
|
)
|
|
|
|
|
2020-04-24 12:14:56 +08:00
|
|
|
return model
|