only keep vit_base_p16.py

mmpretrain/configs/_base_/models/vit_base_p16.py

@@ -2,7 +2,7 @@
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.model.weight_init import KaimingInit
 
-from mmpretrain.models import (CrossEntropyLoss, ImageClassifier,
+from mmpretrain.models import (ImageClassifier, LabelSmoothLoss,
                                VisionTransformer, VisionTransformerClsHead)
 
 # model settings
@@ -10,7 +10,7 @@ model = dict(
     type=ImageClassifier,
     backbone=dict(
         type=VisionTransformer,
-        arch='l',
+        arch='b',
         img_size=224,
         patch_size=16,
         drop_rate=0.1,
@@ -25,7 +25,7 @@ model = dict(
     head=dict(
         type=VisionTransformerClsHead,
         num_classes=1000,
-        in_channels=1024,
+        in_channels=768,
-        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
+        loss=dict(
-        topk=(1, 5),
+            type=LabelSmoothLoss, label_smooth_val=0.1, mode='classy_vision'),
     ))

mmpretrain/configs/_base_/models/vit_base_p32.py

@@ -1,31 +0,0 @@
-# Copyright (c) OpenMMLab. All rights reserved.
-# This is a BETA new format config file, and the usage may change recently.
-from mmengine.model.weight_init import KaimingInit
-
-from mmpretrain.models import (CrossEntropyLoss, ImageClassifier,
-                               VisionTransformer, VisionTransformerClsHead)
-
-# model settings
-model = dict(
-    type=ImageClassifier,
-    backbone=dict(
-        type=VisionTransformer,
-        arch='b',
-        img_size=224,
-        patch_size=32,
-        drop_rate=0.1,
-        init_cfg=[
-            dict(
-                type=KaimingInit,
-                layer='Conv2d',
-                mode='fan_in',
-                nonlinearity='linear')
-        ]),
-    neck=None,
-    head=dict(
-        type=VisionTransformerClsHead,
-        num_classes=1000,
-        in_channels=768,
-        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
-        topk=(1, 5),
-    ))

mmpretrain/configs/_base_/models/vit_large_p16.py

@@ -1,31 +0,0 @@
-# Copyright (c) OpenMMLab. All rights reserved.
-# This is a BETA new format config file, and the usage may change recently.
-from mmengine.model.weight_init import KaimingInit
-
-from mmpretrain.models import (CrossEntropyLoss, ImageClassifier,
-                               VisionTransformer, VisionTransformerClsHead)
-
-# model settings
-model = dict(
-    type=ImageClassifier,
-    backbone=dict(
-        type=VisionTransformer,
-        arch='l',
-        img_size=224,
-        patch_size=16,
-        drop_rate=0.1,
-        init_cfg=[
-            dict(
-                type=KaimingInit,
-                layer='Conv2d',
-                mode='fan_in',
-                nonlinearity='linear')
-        ]),
-    neck=None,
-    head=dict(
-        type=VisionTransformerClsHead,
-        num_classes=1000,
-        in_channels=1024,
-        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
-        topk=(1, 5),
-    ))

mmpretrain/configs/_base_/models/vit_large_p32.py

@@ -1,31 +0,0 @@
-# Copyright (c) OpenMMLab. All rights reserved.
-# This is a BETA new format config file, and the usage may change recently.
-from mmengine.model.weight_init import KaimingInit
-
-from mmpretrain.models import (CrossEntropyLoss, ImageClassifier,
-                               VisionTransformer, VisionTransformerClsHead)
-
-# model settings
-model = dict(
-    type=ImageClassifier,
-    backbone=dict(
-        type=VisionTransformer,
-        arch='l',
-        img_size=224,
-        patch_size=32,
-        drop_rate=0.1,
-        init_cfg=[
-            dict(
-                type=KaimingInit,
-                layer='Conv2d',
-                mode='fan_in',
-                nonlinearity='linear')
-        ]),
-    neck=None,
-    head=dict(
-        type=VisionTransformerClsHead,
-        num_classes=1000,
-        in_channels=1024,
-        loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
-        topk=(1, 5),
-    ))

mmpretrain/configs/vision_transformer/vit_base_p16_32xb128_mae_in1k.py

@@ -1,34 +1,21 @@
 # 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 torch.optim import AdamW
 
 from mmpretrain.engine import EMAHook
-from mmpretrain.models import (ConstantInit, CutMix, ImageClassifier,
+from mmpretrain.models import CutMix, Mixup
-                               LabelSmoothLoss, Mixup, TruncNormalInit,
-                               VisionTransformer, VisionTransformerClsHead)
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_swin_224 import *
     from .._base_.default_runtime import *
+    from .._base_.models.vit_base_p16 import *
     from .._base_.schedules.imagenet_bs1024_adamw_swin import *
 
-# model settings
+model.update(
-model = dict(
+    backbone=dict(drop_rate=0, drop_path_rate=0.1),
-    type=ImageClassifier,
+    head=dict(loss=dict(mode='original')),
-    backbone=dict(
-        type=VisionTransformer,
-        arch='base',
-        img_size=224,
-        patch_size=16,
-        drop_path_rate=0.1),
-    neck=None,
-    head=dict(
-        type=VisionTransformerClsHead,
-        num_classes=1000,
-        in_channels=768,
-        loss=dict(type=LabelSmoothLoss, label_smooth_val=0.1, mode='original'),
-    ),
     init_cfg=[
         dict(type=TruncNormalInit, layer='Linear', std=.02),
         dict(type=ConstantInit, layer='LayerNorm', val=1., bias=0.),
@@ -36,6 +23,7 @@ model = dict(
     train_cfg=dict(
         augments=[dict(type=Mixup, alpha=0.8),
                   dict(type=CutMix, alpha=1.0)]))
+model.backbone.init_cfg = []
 
 # dataset settings
 train_dataloader.update(batch_size=128)

mmpretrain/configs/vision_transformer/vit_base_p32_64xb64_in1k.py

@@ -2,19 +2,26 @@
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
 
-from mmpretrain.models import Mixup
+from mmpretrain.models import CrossEntropyLoss, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_pil_resize_autoaug import *
     from .._base_.default_runtime import *
-    from .._base_.models.vit_base_p32 import *
+    from .._base_.models.vit_base_p16 import *
     from .._base_.schedules.imagenet_bs4096_adamw import *
 
 # model setting
 model.update(
-    head=dict(hidden_dim=3072),
+    backbone=dict(patch_size=32),
+    head=dict(
+        hidden_dim=3072,
+        # loss=dict(type=CrossEntropyLoss, loss_weight=1.0),
+        topk=(1, 5),
+    ),
     train_cfg=dict(augments=dict(type=Mixup, alpha=0.2)),
 )
 
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
 # schedule setting
 optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_base_p32_64xb64_in1k_384px.py

@@ -4,15 +4,19 @@ from mmengine.config import read_base
 
 from mmpretrain.datasets import (CenterCrop, LoadImageFromFile, PackInputs,
                                  RandomFlip, RandomResizedCrop, ResizeEdge)
+from mmpretrain.models import CrossEntropyLoss
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_pil_resize import *
     from .._base_.default_runtime import *
-    from .._base_.models.vit_base_p32 import *
+    from .._base_.models.vit_base_p16 import *
     from .._base_.schedules.imagenet_bs4096_adamw import *
 
 # model setting
-model.update(backbone=dict(img_size=384))
+model.update(
+    backbone=dict(img_size=384, patch_size=32), head=dict(topk=(1, 5)))
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
 
 # dataset setting
 data_preprocessor.update(

mmpretrain/configs/vision_transformer/vit_large_p16_64xb64_in1k.py

@@ -2,19 +2,26 @@
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
 
-from mmpretrain.models import Mixup
+from mmpretrain.models import CrossEntropyLoss, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_pil_resize_autoaug import *
     from .._base_.default_runtime import *
-    from .._base_.models.vit_large_p16 import *
+    from .._base_.models.vit_base_p16 import *
     from .._base_.schedules.imagenet_bs4096_adamw import *
 
 # model setting
 model.update(
-    head=dict(hidden_dim=3072),
+    backbone=dict(arch='l'),
+    head=dict(
+        hidden_dim=3072,
+        in_channels=1024,
+        topk=(1, 5),
+    ),
     train_cfg=dict(augments=dict(type=Mixup, alpha=0.2)),
 )
 
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
 # schedule setting
 optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_large_p16_64xb64_in1k_384px.py

@@ -4,15 +4,20 @@ from mmengine.config import read_base
 
 from mmpretrain.datasets import (CenterCrop, LoadImageFromFile, PackInputs,
                                  RandomFlip, RandomResizedCrop, ResizeEdge)
+from mmpretrain.models import CrossEntropyLoss
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_pil_resize import *
     from .._base_.default_runtime import *
-    from .._base_.models.vit_large_p16 import *
+    from .._base_.models.vit_base_p16 import *
     from .._base_.schedules.imagenet_bs4096_adamw import *
 
 # model setting
-model.update(backbone=dict(img_size=384))
+model.update(
+    backbone=dict(arch='l', img_size=384),
+    head=dict(in_channels=1024, topk=(1, 5)))
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
 
 # dataset setting
 data_preprocessor.update(

mmpretrain/configs/vision_transformer/vit_large_p32_64xb64_in1k.py

@@ -2,19 +2,26 @@
 # This is a BETA new format config file, and the usage may change recently.
 from mmengine.config import read_base
 
-from mmpretrain.models import Mixup
+from mmpretrain.models import CrossEntropyLoss, Mixup
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_pil_resize_autoaug import *
     from .._base_.default_runtime import *
-    from .._base_.models.vit_large_p32 import *
+    from .._base_.models.vit_base_p16 import *
     from .._base_.schedules.imagenet_bs4096_adamw import *
 
 # model setting
 model.update(
-    head=dict(hidden_dim=3072),
+    backbone=dict(arch='l', patch_size=32),
+    head=dict(
+        hidden_dim=3072,
+        in_channels=1024,
+        topk=(1, 5),
+    ),
     train_cfg=dict(augments=dict(type=Mixup, alpha=0.2)),
 )
 
+loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
+
 # schedule setting
 optim_wrapper.update(clip_grad=dict(max_norm=1.0))

mmpretrain/configs/vision_transformer/vit_large_p32_64xb64_in1k_384px.py

@@ -4,18 +4,23 @@ from mmengine.config import read_base
 
 from mmpretrain.datasets import (CenterCrop, LoadImageFromFile, PackInputs,
                                  RandomFlip, RandomResizedCrop, ResizeEdge)
+from mmpretrain.models import CrossEntropyLoss
 
 with read_base():
     from .._base_.datasets.imagenet_bs64_pil_resize import *
     from .._base_.default_runtime import *
-    from .._base_.models.vit_large_p32 import *
+    from .._base_.models.vit_base_p16 import *
     from .._base_.schedules.imagenet_bs4096_adamw import *
 
 # model setting
-model.update(backbone=dict(img_size=384))
+model.update(
+    backbone=dict(arch='l', img_size=384, patch_size=32),
+    head=dict(in_channels=1024, topk=(1, 5)))
+
+model.head.loss = dict(type=CrossEntropyLoss, loss_weight=1.0)
 
 # dataset setting
-data_preprocessor = dict(
+data_preprocessor.update(
     mean=[127.5, 127.5, 127.5],
     std=[127.5, 127.5, 127.5],
     # convert image from BGR to RGB