[Fix] fix swin3d config and readme format

configs/recognition/swin/README.md

@@ -20,18 +20,18 @@ The vision community is witnessing a modeling shift from CNNs to Transformers, w
 
 ### Kinetics-400
 
-| frame sampling strategy |  resolution   | gpus | backbone |   pretrain   | top1 acc | top5 acc |   reference top1 acc   |   reference top1 acc   | testing protocol | inference time(video/s) | gpu_mem(M) | params |   config   |   ckpt   |   log   |
-| :---------------------: | :-----------: | :--: | :------: | :----------: | :------: | :------: | :--------------------: | :--------------------: | :--------------: | :---------------------: | :--------: | :----: | :--------: | :------: | :-----: |
-|         32x2x1          | short-side 320 |  8   |  Swin-T  | ImageNet-1k  |  78.29   |  93.58   | [78.46](https://github.com/SwinTransformer/Video-Swin-Transformer/blob/master/configs/recognition/swin/swin_tiny_patch244_window877_kinetics400_1k.py) | [93.46](https://github.com/SwinTransformer/Video-Swin-Transformer/blob/master/configs/recognition/swin/swin_tiny_patch244_window877_kinetics400_1k.py) | 4 clips x 3 crop |            x            |   21072    | 28.2M  | [config](/configs/recognition/swin/swin-tiny_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-tiny_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-tiny_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-241016b2.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-tiny_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-tiny_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
-|         32x2x1          | short-side 320 |  8   |  Swin-S  | ImageNet-1k  |  80.23   |  94.32   | [80.23](https://github.com/SwinTransformer/Video-Swin-Transformer/blob/master/configs/recognition/swin/swin_small_patch244_window877_kinetics400_1k.py) | [94.16](https://github.com/SwinTransformer/Video-Swin-Transformer/blob/master/configs/recognition/swin/swin_small_patch244_window877_kinetics400_1k.py) | 4 clips x 3 crop |            x            |   33632    | 49.8M  | [config](/configs/recognition/swin/swin-small_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-small_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-small_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-e91ab986.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-small_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-small_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
-|         32x2x1          | short-side 320 |  8   |  Swin-B  | ImageNet-1k  |  80.21   |  94.32   | [80.27](https://github.com/SwinTransformer/Video-Swin-Transformer/blob/master/configs/recognition/swin/swin_base_patch244_window877_kinetics400_1k.py) | [94.42](https://github.com/SwinTransformer/Video-Swin-Transformer/blob/master/configs/recognition/swin/swin_base_patch244_window877_kinetics400_1k.py) | 4 clips x 3 crop |            x            |   45143    | 88.0M  | [config](/configs/recognition/swin/swin-base_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-base_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-base_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-182ec6cc.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-base_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-base_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
-|         32x2x1          | short-side 320 |  8   |  Swin-L  | ImageNet-22k |  83.15   |  95.76   |         83.1\*         |         95.9\*         | 4 clips x 3 crop |            x            |   68881    | 197.0M | [config](/configs/recognition/swin/swin-large_p244-w877-in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large_p244-w877-in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-large_p244-w877-in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-78ad8b11.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large_p244-w877-in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-large_p244-w877-in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
+| frame sampling strategy |   resolution   | gpus | backbone |   pretrain   | top1 acc | top5 acc |    reference top1 acc     |    reference top1 acc     | testing protocol  | gpu_mem(M) | FLOPs | params |    config     |    ckpt     |     log     |
+| :---------------------: | :------------: | :--: | :------: | :----------: | :------: | :------: | :-----------------------: | :-----------------------: | :---------------: | :--------: | :---: | :----: | :-----------: | :---------: | :---------: |
+|         32x2x1          | short-side 320 |  8   |  Swin-T  | ImageNet-1k  |  78.29   |  93.58   | 78.46 \[[VideoSwin](https://github.com/SwinTransformer/Video-Swin-Transformer)\] | 93.46 \[[VideoSwin](https://github.com/SwinTransformer/Video-Swin-Transformer)\] | 4 clips x 3 crops |   21072    |  88G  | 28.2M  | [config](/configs/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-241016b2.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
+|         32x2x1          | short-side 320 |  8   |  Swin-S  | ImageNet-1k  |  80.23   |  94.32   | 80.23 \[[VideoSwin](https://github.com/SwinTransformer/Video-Swin-Transformer)\] | 94.16 \[[VideoSwin](https://github.com/SwinTransformer/Video-Swin-Transformer)\] | 4 clips x 3 crops |   33632    | 166G  | 49.8M  | [config](/configs/recognition/swin/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-e91ab986.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
+|         32x2x1          | short-side 320 |  8   |  Swin-B  | ImageNet-1k  |  80.21   |  94.32   | 80.27 \[[VideoSwin](https://github.com/SwinTransformer/Video-Swin-Transformer)\] | 94.42 \[[VideoSwin](https://github.com/SwinTransformer/Video-Swin-Transformer)\] | 4 clips x 3 crops |   45143    | 282G  | 88.0M  | [config](/configs/recognition/swin/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-182ec6cc.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
+|         32x2x1          | short-side 320 |  8   |  Swin-L  | ImageNet-22k |  83.15   |  95.76   |          83.1\*           |          95.9\*           | 4 clips x 3 crops |   68881    | 604G  |  197M  | [config](/configs/recognition/swin/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-78ad8b11.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log) |
 
 ### Kinetics-700
 
-| frame sampling strategy |   resolution   | gpus | backbone |   pretrain   | top1 acc | top5 acc | testing protocol | inference time(video/s) | gpu_mem(M) | params |          config          |          ckpt          |          log           |
-| :---------------------: | :------------: | :--: | :------: | :----------: | :------: | :------: | :--------------: | :---------------------: | :--------: | :----: | :----------------------: | :--------------------: | :--------------------: |
-|         32x2x1          | short-side 320 |  16  |  Swin-L  | ImageNet-22k |  75.26   |  92.44   | 4 clips x 3 crop |            x            |   68898    | 197.4M | [config](/configs/recognition/swin/swin-large_p244-w877-in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large_p244-w877-in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb/swin-large_p244-w877-in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb_20220930-f8d74db7.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large_p244-w877-in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb/swin-large_p244-w877-in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.py.log) |
+| frame sampling strategy |   resolution   | gpus | backbone |   pretrain   | top1 acc | top5 acc | testing protocol  | gpu_mem(M) | FLOPs | params |             config             |             ckpt             |             log             |
+| :---------------------: | :------------: | :--: | :------: | :----------: | :------: | :------: | :---------------: | :--------: | :---: | :----: | :----------------------------: | :--------------------------: | :-------------------------: |
+|         32x2x1          | short-side 320 |  16  |  Swin-L  | ImageNet-22k |  75.26   |  92.44   | 4 clips x 3 crops |   68898    | 604G  |  197M  | [config](/configs/recognition/swin/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb_20220930-f8d74db7.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.py.log) |
 
 1. The **gpus** indicates the number of gpus we used to get the checkpoint. If you want to use a different number of gpus or videos per gpu, the best way is to set `--auto-scale-lr` when calling `tools/train.py`, this parameter will auto-scale the learning rate according to the actual batch size and the original batch size.
 2. The values in columns named after "reference" are the results got by testing on our dataset, using the checkpoints provided by the author with same model settings. `*` means that the numbers are copied from the paper.
@@ -51,7 +51,7 @@ python tools/train.py ${CONFIG_FILE} [optional arguments]
 Example: train VideoSwin model on Kinetics-400 dataset in a deterministic option with periodic validation.
 
 ```shell
-python tools/train.py configs/recognition/swin/swin-tiny_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py \
+python tools/train.py configs/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py \
     --cfg-options randomness.seed=0 randomness.deterministic=True
 ```
 
@@ -68,7 +68,7 @@ python tools/test.py ${CONFIG_FILE} ${CHECKPOINT_FILE} [optional arguments]
 Example: test VideoSwin model on Kinetics-400 dataset and dump the result to a pkl file.
 
 ```shell
-python tools/test.py configs/recognition/swin/swin-tiny_p244-w877-in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py \
+python tools/test.py configs/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py \
     checkpoints/SOME_CHECKPOINT.pth --dump result.pkl
 ```
 

configs/recognition/swin/metafile.yml

@@ -0,0 +1,122 @@
+Collections:
+- Name: Swin
+  README: configs/recognition/swin/README.md
+  Paper:
+    URL: https://arxiv.org/abs/2106.13230
+    Title: 'Video Swin Transformer'
+
+Models:
+  - Name: swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb
+    Config: configs/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py
+    In Collection: Swin
+    Metadata:
+      Architecture: Swin-T
+      Batch Size: 8
+      Epochs: 30
+      FLOPs: 88G
+      Parameters: 28.2M
+      Pretrained: ImageNet-1K
+      Resolution: short-side 320
+      Training Data: Kinetics-400
+      Training Resources: 8 GPUs
+    Modality: RGB
+    Results:
+      - Dataset: Kinetics-400
+        Task: Action Recognition
+        Metrics:
+              Top 1 Accuracy: 78.29
+              Top 5 Accuracy: 93.58
+    Training Log: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log
+    Weights: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-tiny-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-241016b2.pth
+
+  - Name: swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb
+    Config: configs/recognition/swin/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py
+    In Collection: Swin
+    Metadata:
+      Architecture: Swin-S
+      Batch Size: 8
+      Epochs: 30
+      FLOPs: 166G
+      Parameters: 49.8M
+      Pretrained: ImageNet-1K
+      Resolution: short-side 320
+      Training Data: Kinetics-400
+      Training Resources: 8 GPUs
+    Modality: RGB
+    Results:
+    - Dataset: Kinetics-400
+      Task: Action Recognition
+      Metrics:
+        Top 1 Accuracy: 80.23
+        Top 5 Accuracy: 94.32
+    Training Log: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log
+    Weights: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-small-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-e91ab986.pth
+
+  - Name: swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb
+    Config: configs/recognition/swin/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py.py
+    In Collection: Swin
+    Metadata:
+      Architecture: Swin-B
+      Batch Size: 8
+      Epochs: 30
+      FLOPs: 282G
+      Parameters: 88.0M
+      Pretrained: ImageNet-1K
+      Resolution: short-side 320
+      Training Data: Kinetics-400
+      Training Resources: 8 GPUs
+    Modality: RGB
+    Results:
+    - Dataset: Kinetics-400
+      Task: Action Recognition
+      Metrics:
+        Top 1 Accuracy: 80.21
+        Top 5 Accuracy: 94.32
+    Training Log: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log
+    Weights: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-base-p244-w877_in1k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-182ec6cc.pth
+
+  - Name: swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb
+    Config: configs/recognition/swin/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py
+    In Collection: Swin
+    Metadata:
+      Architecture: Swin-L
+      Batch Size: 8
+      Epochs: 30
+      FLOPs: 604G
+      Parameters: 197M
+      Pretrained: ImageNet-22K
+      Resolution: short-side 320
+      Training Data: Kinetics-400
+      Training Resources: 8 GPUs
+    Modality: RGB
+    Results:
+    - Dataset: Kinetics-400
+      Task: Action Recognition
+      Metrics:
+        Top 1 Accuracy: 83.15
+        Top 5 Accuracy: 95.76
+    Training Log: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.log
+    Weights: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb/swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb_20220930-78ad8b11.pth
+
+  - Name: swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb
+    Config: configs/recognition/swin/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.py
+    In Collection: Swin
+    Metadata:
+      Architecture: Swin-L
+      Batch Size: 8
+      Epochs: 30
+      FLOPs: 604G
+      Parameters: 197M
+      Pretrained: ImageNet-22K
+      Resolution: short-side 320
+      Training Data: Kinetics-700
+      Training Resources: 16 GPUs
+    Modality: RGB
+    Results:
+    - Dataset: Kinetics-700
+      Task: Action Recognition
+      Metrics:
+        Top 1 Accuracy: 75.26
+        Top 5 Accuracy: 92.44
+    Training Log: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.log
+    Weights: https://download.openmmlab.com/mmaction/v1.0/recognition/swin/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb_20220930-f8d74db7.pth

configs/recognition/swin/swin-large_p244-w877-in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.py → configs/recognition/swin/swin-large-p244-w877_in22k-pre_16xb8-amp-32x2x1-30e_kinetics700-rgb.py

@@ -1,5 +1,5 @@
 _base_ = [
-    'swin-large_p244-w877-in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py'
+    'swin-large-p244-w877_in22k-pre_8xb8-amp-32x2x1-30e_kinetics400-rgb.py'
 ]
 
 model = dict(cls_head=dict(num_classes=700))
@@ -15,7 +15,7 @@
 #     io_backend='petrel',
 #     path_mapping=dict(
 #         {'data/kinetics700': 's3://openmmlab/datasets/action/Kinetics700'}))
-file_client_args = dict(backend='disk')
+file_client_args = dict(io_backend='disk')
 train_pipeline = [
     dict(type='DecordInit', **file_client_args),
     dict(type='SampleFrames', clip_len=32, frame_interval=2, num_clips=1),

model-index.yml

@@ -13,6 +13,7 @@ Import:
 - configs/recognition/tanet/metafile.yml
 - configs/recognition/x3d/metafile.yml
 - configs/recognition/trn/metafile.yml
+- configs/recognition/swin/metafile.yml
 - configs/detection/ava/metafile.yml
 - configs/detection/acrn/metafile.yml
 - configs/skeleton/stgcn/metafile.yml

tests/models/recognizers/test_recognizer3d.py

@@ -115,3 +115,13 @@ def test_tpn_slowonly():
     loss_vars, _ = train_test_step(config, input_shape=input_shape)
     assert 'loss_aux' in loss_vars
     assert loss_vars['loss_cls'] + loss_vars['loss_aux'] == loss_vars['loss']
+
+
+def test_swin():
+    register_all_modules()
+    config = get_recognizer_cfg('swin/swin-tiny-p244-w877_in1k-pre_'
+                                '8xb8-amp-32x2x1-30e_kinetics400-rgb.py')
+    config.model['backbone']['pretrained2d'] = False
+    config.model['backbone']['pretrained'] = None
+    input_shape = (1, 3, 4, 64, 64)  # M C T H W
+    train_test_step(config, input_shape=input_shape)