set arch etc

mmpretrain/configs/swin_transformer/swin_large_8xb8_cub_384px.py

@@ -16,12 +16,12 @@ with read_base():
 # model settings
 checkpoint = 'https://download.openmmlab.com/mmclassification/v0/swin-transformer/convert/swin-large_3rdparty_in21k-384px.pth'  # noqa
 
-model = dict(
-    type=ImageClassifier,
+model.update(
     backbone=dict(
+        arch='large',
         init_cfg=dict(
             type=PretrainedInit, checkpoint=checkpoint, prefix='backbone')),
-    head=dict(num_classes=200, ))
+    head=dict(num_classes=200, in_channels=1536))
 
 # schedule settings
 optim_wrapper = dict(

mmpretrain/configs/swin_transformer_v2/swinv2_base_w12_8xb128_in21k_192px.py

@@ -1,6 +1,9 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet21k_bs128 import *
@@ -9,10 +12,17 @@ with read_base():
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
 # model settings
-model = dict(
-    backbone=dict(img_size=192, window_size=[12, 12, 12, 6]),
+model.update(
+    backbone=dict(
+        img_size=192, drop_path_rate=0.5, window_size=[12, 12, 12, 6]),
     head=dict(num_classes=21841),
-)
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))
 
 # dataset settings
 data_preprocessor = dict(num_classes=21841)

mmpretrain/configs/swin_transformer_v2/swinv2_base_w16_16xb64_in1k_256px.py

@@ -1,6 +1,9 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_256 import *
@@ -8,4 +11,14 @@ with read_base():
     from .._base_.models.swin_transformer_v2_base import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-model = dict(backbone=dict(window_size=[16, 16, 16, 8]))
+# model settings
+model.update(
+    backbone=dict(
+        img_size=256, drop_path_rate=0.5, window_size=[16, 16, 16, 8]),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))

mmpretrain/configs/swin_transformer_v2/swinv2_base_w16_in21k_pre_16xb64_in1k_256px.py

@@ -1,8 +1,9 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
 
-from mmpretrain.models import ImageClassifier
+from mmpretrain.models import CutMix, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_256 import *
@@ -10,9 +11,16 @@ with read_base():
     from .._base_.models.swin_transformer_v2_base import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-model = dict(
-    type=ImageClassifier,
+# model settings
+model.update(
     backbone=dict(
+        img_size=256,
         window_size=[16, 16, 16, 8],
-        drop_path_rate=0.2,
-        pretrained_window_sizes=[12, 12, 12, 6]))
+        pretrained_window_sizes=[12, 12, 12, 6]),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))

mmpretrain/configs/swin_transformer_v2/swinv2_base_w24_in21k_pre_16xb64_in1k_384px.py

@@ -2,18 +2,13 @@
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
 
-from mmpretrain.models import ImageClassifier
-
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_384 import *
     from .._base_.default_runtime import *
     from .._base_.models.swin_transformer_v2_base import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-model = dict(
-    type=ImageClassifier,
+# model settings
+model.update(
     backbone=dict(
-        img_size=384,
-        window_size=[24, 24, 24, 12],
-        drop_path_rate=0.2,
-        pretrained_window_sizes=[12, 12, 12, 6]))
+        window_size=[24, 24, 24, 12], pretrained_window_sizes=[12, 12, 12, 6]))

mmpretrain/configs/swin_transformer_v2/swinv2_large_w12_8xb128_in21k_192px.py

@@ -1,6 +1,9 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet21k_bs128 import *
@@ -9,10 +12,17 @@ with read_base():
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
 # model settings
-model = dict(
-    backbone=dict(img_size=192, window_size=[12, 12, 12, 6]),
+model.update(
+    backbone=dict(
+        img_size=192, drop_path_rate=0.5, window_size=[12, 12, 12, 6]),
     head=dict(num_classes=21841),
-)
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))
 
 # dataset settings
 data_preprocessor = dict(num_classes=21841)

mmpretrain/configs/swin_transformer_v2/swinv2_large_w16_in21k_pre_16xb64_in1k_256px.py

@@ -3,7 +3,7 @@
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
 
-from mmpretrain.models import ImageClassifier
+from mmpretrain.models import CrossEntropyLoss
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_256 import *
@@ -11,8 +11,14 @@ with read_base():
     from .._base_.models.swin_transformer_v2_base import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-model = dict(
-    type=ImageClassifier,
+# model settings
+model.update(
     backbone=dict(
-        window_size=[16, 16, 16, 8], pretrained_window_sizes=[12, 12, 12, 6]),
-)
+        arch='large',
+        img_size=256,
+        window_size=[16, 16, 16, 8],
+        pretrained_window_sizes=[12, 12, 12, 6]),
+    head=dict(
+        in_channels=1536,
+        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
+        topk=(1, 5)))

mmpretrain/configs/swin_transformer_v2/swinv2_large_w24_in21k_pre_16xb64_in1k_384px.py

@@ -3,7 +3,7 @@
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
 
-from mmpretrain.models import ImageClassifier
+from mmpretrain.models import CrossEntropyLoss
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_384 import *
@@ -11,10 +11,14 @@ with read_base():
     from .._base_.models.swin_transformer_v2_base import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-model = dict(
-    type=ImageClassifier,
+# model settings
+model.update(
     backbone=dict(
+        arch='large',
         img_size=384,
         window_size=[24, 24, 24, 12],
         pretrained_window_sizes=[12, 12, 12, 6]),
-)
+    head=dict(
+        in_channels=1536,
+        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
+        topk=(1, 5)))

mmpretrain/configs/swin_transformer_v2/swinv2_small_w16_16xb64_in1k_256px.py

@@ -1,6 +1,9 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_256 import *
@@ -8,4 +11,18 @@ with read_base():
     from .._base_.models.swin_transformer_v2_base import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-model = dict(backbone=dict(window_size=[16, 16, 16, 8]))
+# model settings
+model.update(
+    backbone=dict(
+        arch='small',
+        img_size=256,
+        drop_path_rate=0.3,
+        window_size=[16, 16, 16, 8]),
+    head=dict(in_channels=768),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))

mmpretrain/configs/swin_transformer_v2/swinv2_tiny_w16_16xb64_in1k_256px.py

@@ -1,6 +1,9 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
+from mmengine.model import ConstantInit, TruncNormalInit
+
+from mmpretrain.models import CutMix, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_256 import *
@@ -8,4 +11,18 @@ with read_base():
     from .._base_.models.swin_transformer_v2_base import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-model = dict(backbone=dict(window_size=[16, 16, 16, 8]))
+# model settings
+model.update(
+    backbone=dict(
+        arch='tiny',
+        img_size=256,
+        drop_path_rate=0.2,
+        window_size=[16, 16, 16, 8]),
+    head=dict(in_channels=768),
+    init_cfg=[
+        dict(type=TruncNormalInit, layer='Linear', std=0.02, bias=0.),
+        dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.)
+    ],
+    train_cfg=dict(
+        augments=[dict(type=Mixup, alpha=0.8),
+                  dict(type=CutMix, alpha=1.0)]))