Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[Feature] Support using backbones from pytorch-image-models (timm) for TSN #880

Merged
merged 10 commits into from
Jun 10, 2021
Merged

[Feature] Support using backbones from pytorch-image-models (timm) for TSN #880

merged 10 commits into from
Jun 10, 2021

Conversation

irvingzhang0512
Copy link
Contributor

@irvingzhang0512 irvingzhang0512 commented May 22, 2021 •
edited
Loading

Motivation

pytorch-image-models is one of the best pytorch classification projects, with a large number of backbones, including both cnns and transformers. It would be interesting if we can train tsn models (and other 2d recognizers) with various backbones.

Modification

Following #679 #720, add codes, docs, unittest.

Use cases (Optional)

Modify model settings, an example is listed as follows.

# model settings
model = dict(
    type='Recognizer2D',
    backbone=dict(type='timm.swin_base_patch4_window7_224', pretrained=True),
    cls_head=dict(
        type='TSNHead',
        num_classes=400,
        in_channels=1024,
        spatial_type='avg',
        consensus=dict(type='AvgConsensus', dim=1),
        dropout_ratio=0.4,
        init_std=0.01),
    # model training and testing settings
    train_cfg=None,
    test_cfg=dict(average_clips=None))
  • There is a factory function for timm and we create timm backbone by timm.create_model(backbone_type, **backbone)

https://github.com/rwightman/pytorch-image-models/blob/23c18a33e4168dc7cb11439c1f9acd38dc8e9824/timm/models/factory.py#L25

  • For different models, users should modify model['cls_head']['in_channels'] and img_norm_cfg accordingly.

Test timm backbones

There are over 500 backbones in timm...... Write a script to test timm backbones.

check_timm_models.py.txt

Please note that

  1. backbones with verify=True are supported for MMAction2 TSN models.
  2. mean&std columns are designed for pixels in [0, 1] floats. So if you want to use this mean/std in MMAction2 img_norm_cfg, both std and mean should be multiplied by 255.
model name input size mean std in channels pretrained verify
adv_inception_v3 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
botnet26t_256 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
botnet50ts_256 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
cait_m36_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True True
cait_m48_448 (3, 448, 448) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True MMAction2 Test Error
cait_s24_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 384 True True
cait_s24_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 384 True True
cait_s36_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 384 True True
cait_xs24_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 288 True True
cait_xxs24_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 192 True True
cait_xxs24_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 192 True True
cait_xxs36_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 192 True True
cait_xxs36_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 192 True True
coat_lite_mini (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
coat_lite_small (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
coat_lite_tiny (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 320 True True
coat_mini (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 216 False True
coat_tiny (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 152 False True
cspdarknet53 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
cspdarknet53_iabn (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
cspresnet50 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
cspresnet50d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
cspresnet50w (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
cspresnext50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
cspresnext50_iabn (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
darknet53 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
densenet121 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
densenet121d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
densenet161 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2208 True True
densenet169 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1664 True True
densenet201 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1920 True True
densenet264 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4032 False True
densenet264d_iabn (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
densenetblur121d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla34 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
dla46_c (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 256 True True
dla46x_c (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 256 True True
dla60 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla60_res2net (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla60_res2next (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla60x (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla60x_c (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 256 True True
dla102 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla102x (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla102x2 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dla169 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
dm_nfnet_f0 (3, 192, 192) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
dm_nfnet_f1 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
dm_nfnet_f2 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
dm_nfnet_f3 (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
dm_nfnet_f4 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
dm_nfnet_f5 (3, 416, 416) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
dm_nfnet_f6 (3, 448, 448) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
dpn68 (3, 224, 224) (0.486, 0.459, 0.408) (0.235, 0.235, 0.235) 832 True True
dpn68b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 832 True True
dpn92 (3, 224, 224) (0.486, 0.459, 0.408) (0.235, 0.235, 0.235) 2688 True True
dpn98 (3, 224, 224) (0.486, 0.459, 0.408) (0.235, 0.235, 0.235) 2688 True True
dpn107 (3, 224, 224) (0.486, 0.459, 0.408) (0.235, 0.235, 0.235) 2688 True True
dpn131 (3, 224, 224) (0.486, 0.459, 0.408) (0.235, 0.235, 0.235) 2688 True True
eca_nfnet_l0 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2304 True True
eca_nfnet_l1 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 True True
eca_vovnet39b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
ecaresnet26t (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ecaresnet50d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ecaresnet50d_pruned (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2022 True True
ecaresnet50t (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ecaresnet101d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ecaresnet101d_pruned (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2042 True True
ecaresnet200d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
ecaresnet269d (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ecaresnetlight (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ecaresnext26t_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
ecaresnext50t_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
efficientnet_b0 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
efficientnet_b1 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
efficientnet_b1_pruned (3, 240, 240) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
efficientnet_b2 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1408 True True
efficientnet_b2_pruned (3, 260, 260) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1408 True True
efficientnet_b2a (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1408 True True
efficientnet_b3 (3, 288, 288) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
efficientnet_b3_pruned (3, 300, 300) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1536 True True
efficientnet_b3a (3, 288, 288) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
efficientnet_b4 (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1792 True True
efficientnet_b5 (3, 456, 456) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
efficientnet_b6 (3, 528, 528) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2304 False True
efficientnet_b7 (3, 600, 600) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 False True
efficientnet_b8 (3, 672, 672) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2816 False True
efficientnet_cc_b0_4e (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnet_cc_b0_8e (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnet_cc_b1_8e (3, 240, 240) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnet_el (3, 300, 300) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
efficientnet_el_pruned (3, 300, 300) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
efficientnet_em (3, 240, 240) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
efficientnet_es (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
efficientnet_es_pruned (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
efficientnet_l2 (3, 800, 800) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 5504 False MMAction2 Test Error
efficientnet_lite0 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
efficientnet_lite1 (3, 240, 240) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnet_lite2 (3, 260, 260) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnet_lite3 (3, 300, 300) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnet_lite4 (3, 380, 380) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnetv2_l (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False MMAction2 Test Error
efficientnetv2_m (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
efficientnetv2_rw_s (3, 288, 288) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1792 True True
efficientnetv2_s (3, 288, 288) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
ens_adv_inception_resnet_v2 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1536 True True
ese_vovnet19b_dw (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
ese_vovnet19b_slim (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
ese_vovnet19b_slim_dw (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
ese_vovnet39b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
ese_vovnet39b_evos (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
ese_vovnet57b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
ese_vovnet99b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
ese_vovnet99b_iabn (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
fbnetc_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1984 True True
gernet_l (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True True
gernet_m (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True True
gernet_s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1920 True True
ghostnet_050 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
ghostnet_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
ghostnet_130 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
gluon_inception_v3 (3, 299, 299) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet18_v1b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
gluon_resnet34_v1b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
gluon_resnet50_v1b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet50_v1c (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet50_v1d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet50_v1s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet101_v1b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet101_v1c (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet101_v1d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet101_v1s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet152_v1b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet152_v1c (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet152_v1d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnet152_v1s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnext101_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_resnext101_64x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_senet154 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_seresnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_seresnext101_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_seresnext101_64x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
gluon_xception65 (3, 299, 299) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
halonet26t (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
halonet50ts (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
halonet_h1 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
halonet_h1_c4c5 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
hardcorenas_a (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
hardcorenas_b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
hardcorenas_c (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
hardcorenas_d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
hardcorenas_e (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
hardcorenas_f (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
hrnet_w18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w18_small (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w18_small_v2 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w30 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w32 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w40 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w44 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w48 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
hrnet_w64 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ig_resnext101_32x8d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ig_resnext101_32x16d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ig_resnext101_32x32d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True MMAction2 Test Error
ig_resnext101_32x48d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True MMAction2 Test Error
inception_resnet_v2 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1536 True True
inception_v3 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
inception_v4 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1536 True True
lambda_resnet26t (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
lambda_resnet50t (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
legacy_senet154 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
legacy_seresnet18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
legacy_seresnet34 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
legacy_seresnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
legacy_seresnet101 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
legacy_seresnet152 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
legacy_seresnext26_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
legacy_seresnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
legacy_seresnext101_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
mixer_b16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / True Function 'forward_features' doesn't exist
mixer_b16_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / True Function 'forward_features' doesn't exist
mixer_b32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / False Function 'forward_features' doesn't exist
mixer_l16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / True Function 'forward_features' doesn't exist
mixer_l16_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / True Function 'forward_features' doesn't exist
mixer_l32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / False Function 'forward_features' doesn't exist
mixer_s16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / False Function 'forward_features' doesn't exist
mixer_s16_glu_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / False Function 'forward_features' doesn't exist
mixer_s32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / False Function 'forward_features' doesn't exist
mixnet_l (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
mixnet_m (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
mixnet_s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
mixnet_xl (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
mixnet_xxl (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 False True
mnasnet_050 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
mnasnet_075 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
mnasnet_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mnasnet_140 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
mnasnet_a1 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mnasnet_b1 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mnasnet_small (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
mobilenetv2_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mobilenetv2_110d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mobilenetv2_120d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mobilenetv2_140 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1792 True True
mobilenetv3_large_075 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
mobilenetv3_large_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mobilenetv3_large_100_miil (3, 224, 224) (0, 0, 0) (1, 1, 1) 1280 True True
mobilenetv3_large_100_miil_in21k (3, 224, 224) (0, 0, 0) (1, 1, 1) 1280 True True
mobilenetv3_rw (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
mobilenetv3_small_075 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
mobilenetv3_small_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
nasnetalarge (3, 331, 331) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 4032 True True
nf_ecaresnet26 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nf_ecaresnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nf_ecaresnet101 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nf_regnet_b0 (3, 192, 192) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 960 False True
nf_regnet_b1 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 960 True True
nf_regnet_b2 (3, 240, 240) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1064 False True
nf_regnet_b3 (3, 288, 288) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1152 False True
nf_regnet_b4 (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1344 False True
nf_regnet_b5 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 False True
nf_resnet26 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nf_resnet50 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
nf_resnet101 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nf_seresnet26 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nf_seresnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nf_seresnet101 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
nfnet_f0 (3, 192, 192) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False True
nfnet_f0s (3, 192, 192) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False True
nfnet_f1 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False True
nfnet_f1s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False True
nfnet_f2 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False True
nfnet_f2s (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False True
nfnet_f3 (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f3s (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f4 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f4s (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f5 (3, 416, 416) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f5s (3, 416, 416) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f6 (3, 448, 448) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f6s (3, 448, 448) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f7 (3, 480, 480) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_f7s (3, 480, 480) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3072 False MMAction2 Test Error
nfnet_l0 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2304 True True
pit_b_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1 True MMAction2 Test Error
pit_b_distilled_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2 True MMAction2 Test Error
pit_s_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1 True MMAction2 Test Error
pit_s_distilled_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2 True MMAction2 Test Error
pit_ti_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1 True MMAction2 Test Error
pit_ti_distilled_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2 True MMAction2 Test Error
pit_xs_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1 True MMAction2 Test Error
pit_xs_distilled_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2 True MMAction2 Test Error
pnasnet5large (3, 331, 331) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 4320 True MMAction2 Test Error
regnetx_002 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 368 True True
regnetx_004 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 384 True True
regnetx_006 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 528 True True
regnetx_008 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 672 True True
regnetx_016 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 912 True True
regnetx_032 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1008 True True
regnetx_040 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1360 True True
regnetx_064 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1624 True True
regnetx_080 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1920 True True
regnetx_120 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2240 True True
regnetx_160 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
regnetx_320 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2520 True True
regnety_002 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 368 True True
regnety_004 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 440 True True
regnety_006 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 608 True True
regnety_008 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True True
regnety_016 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 888 True True
regnety_032 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1512 True True
regnety_040 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1088 True True
regnety_064 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1296 True True
regnety_080 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2016 True True
regnety_120 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2240 True True
regnety_160 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3024 True True
regnety_320 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 3712 True True
repvgg_a2 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1408 True True
repvgg_b0 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
repvgg_b1 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
repvgg_b1g4 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
repvgg_b2 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True True
repvgg_b2g4 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True True
repvgg_b3 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True True
repvgg_b3g4 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True True
res2net50_14w_8s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
res2net50_26w_4s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
res2net50_26w_6s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
res2net50_26w_8s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
res2net50_48w_2s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
res2net101_26w_4s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
res2next50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnest14d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnest26d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnest50d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnest50d_1s4x24d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnest50d_4s2x40d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnest101e (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnest200e (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnest269e (3, 416, 416) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
resnet18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
resnet18d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
resnet26 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnet26d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnet26t (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
resnet34 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
resnet34d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
resnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnet50d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnet50t (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
resnet101 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
resnet101d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnet152 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
resnet152d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnet200 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
resnet200d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetblur18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
resnetblur50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetrs50 (3, 160, 160) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetrs101 (3, 192, 192) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetrs152 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetrs200 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetrs270 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetrs350 (3, 288, 288) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnetrs420 (3, 320, 320) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True MMAction2 Test Error
resnetv2_50x1_bitm (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
resnetv2_50x1_bitm_in21k (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
resnetv2_50x3_bitm (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 6144 True MMAction2 Test Error
resnetv2_50x3_bitm_in21k (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 6144 True MMAction2 Test Error
resnetv2_101x1_bitm (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
resnetv2_101x1_bitm_in21k (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
resnetv2_101x3_bitm (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 6144 True MMAction2 Test Error
resnetv2_101x3_bitm_in21k (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 6144 True MMAction2 Test Error
resnetv2_152x2_bitm (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 4096 True MMAction2 Test Error
resnetv2_152x2_bitm_in21k (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 4096 True MMAction2 Test Error
resnetv2_152x4_bitm (3, 480, 480) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) / True Model Inference Error
resnetv2_152x4_bitm_in21k / / / / / Model Init Error
resnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnext50d_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnext101_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
resnext101_32x8d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
resnext101_64x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
rexnet_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
rexnet_130 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
rexnet_150 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
rexnet_200 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
rexnetr_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
rexnetr_130 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
rexnetr_150 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
rexnetr_200 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
selecsls42 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
selecsls42b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
selecsls60 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
selecsls60b (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
selecsls84 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
semnasnet_050 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
semnasnet_075 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
semnasnet_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
semnasnet_140 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 False True
senet154 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
seresnet18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
seresnet34 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 False True
seresnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
seresnet50t (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
seresnet101 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
seresnet152 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
seresnet152d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
seresnet200d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
seresnet269d (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
seresnext26d_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
seresnext26t_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
seresnext26tn_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
seresnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
seresnext101_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
seresnext101_32x8d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
skresnet18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
skresnet34 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
skresnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
skresnet50d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / False Model Inference Error
skresnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
spnasnet_100 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
ssl_resnet18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
ssl_resnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ssl_resnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ssl_resnext101_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ssl_resnext101_32x8d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
ssl_resnext101_32x16d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
swin_base_patch4_window7_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
swin_base_patch4_window7_224_in22k (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
swin_base_patch4_window12_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
swin_base_patch4_window12_384_in22k (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
swin_large_patch4_window7_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
swin_large_patch4_window7_224_in22k (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
swin_large_patch4_window12_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True MMAction2 Test Error
swin_large_patch4_window12_384_in22k (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True MMAction2 Test Error
swin_small_patch4_window7_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True True
swin_tiny_patch4_window7_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True True
swinnet26t_256 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
swinnet50ts_256 (3, 256, 256) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 False True
swsl_resnet18 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
swsl_resnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
swsl_resnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
swsl_resnext101_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
swsl_resnext101_32x8d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
swsl_resnext101_32x16d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
tf_efficientnet_b0 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
tf_efficientnet_b0_ap (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_b0_ns (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
tf_efficientnet_b1 (3, 240, 240) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
tf_efficientnet_b1_ap (3, 240, 240) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_b1_ns (3, 240, 240) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
tf_efficientnet_b2 (3, 260, 260) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1408 True True
tf_efficientnet_b2_ap (3, 260, 260) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1408 True True
tf_efficientnet_b2_ns (3, 260, 260) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1408 True True
tf_efficientnet_b3 (3, 300, 300) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
tf_efficientnet_b3_ap (3, 300, 300) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1536 True True
tf_efficientnet_b3_ns (3, 300, 300) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
tf_efficientnet_b4 (3, 380, 380) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1792 True True
tf_efficientnet_b4_ap (3, 380, 380) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1792 True True
tf_efficientnet_b4_ns (3, 380, 380) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1792 True True
tf_efficientnet_b5 (3, 456, 456) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True MMAction2 Test Error
tf_efficientnet_b5_ap (3, 456, 456) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True MMAction2 Test Error
tf_efficientnet_b5_ns (3, 456, 456) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True MMAction2 Test Error
tf_efficientnet_b6 (3, 528, 528) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2304 True MMAction2 Test Error
tf_efficientnet_b6_ap (3, 528, 528) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2304 True MMAction2 Test Error
tf_efficientnet_b6_ns (3, 528, 528) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2304 True MMAction2 Test Error
tf_efficientnet_b7 (3, 600, 600) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True MMAction2 Test Error
tf_efficientnet_b7_ap (3, 600, 600) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2560 True MMAction2 Test Error
tf_efficientnet_b7_ns (3, 600, 600) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2560 True MMAction2 Test Error
tf_efficientnet_b8 (3, 672, 672) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2816 True MMAction2 Test Error
tf_efficientnet_b8_ap (3, 672, 672) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2816 True MMAction2 Test Error
tf_efficientnet_cc_b0_4e (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_cc_b0_8e (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_cc_b1_8e (3, 240, 240) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_el (3, 300, 300) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1536 True True
tf_efficientnet_em (3, 240, 240) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_es (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_l2_ns (3, 800, 800) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 5504 True MMAction2 Test Error
tf_efficientnet_l2_ns_475 (3, 475, 475) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 5504 True MMAction2 Test Error
tf_efficientnet_lite0 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_lite1 (3, 240, 240) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_lite2 (3, 260, 260) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_lite3 (3, 300, 300) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnet_lite4 (3, 380, 380) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnetv2_b0 (3, 192, 192) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
tf_efficientnetv2_b1 (3, 192, 192) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1280 True True
tf_efficientnetv2_b2 (3, 208, 208) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1408 True True
tf_efficientnetv2_b3 (3, 240, 240) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
tf_efficientnetv2_l (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True MMAction2 Test Error
tf_efficientnetv2_l_in21ft1k (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True MMAction2 Test Error
tf_efficientnetv2_l_in21k (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True MMAction2 Test Error
tf_efficientnetv2_m (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnetv2_m_in21ft1k (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnetv2_m_in21k (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnetv2_s (3, 300, 300) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnetv2_s_in21ft1k (3, 300, 300) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_efficientnetv2_s_in21k (3, 300, 300) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_inception_v3 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
tf_mixnet_l (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
tf_mixnet_m (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
tf_mixnet_s (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1536 True True
tf_mobilenetv3_large_075 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_mobilenetv3_large_100 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_mobilenetv3_large_minimal_100 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 True True
tf_mobilenetv3_small_075 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True True
tf_mobilenetv3_small_100 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True True
tf_mobilenetv3_small_minimal_100 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True True
tnt_b_patch16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 640 False True
tnt_s_patch16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 384 True True
tresnet_l (3, 224, 224) (0, 0, 0) (1, 1, 1) / True Model Inference Error
tresnet_l_448 (3, 448, 448) (0, 0, 0) (1, 1, 1) / True Model Inference Error
tresnet_m (3, 224, 224) (0, 0, 0) (1, 1, 1) / True Model Inference Error
tresnet_m_448 (3, 448, 448) (0, 0, 0) (1, 1, 1) / True Model Inference Error
tresnet_m_miil_in21k (3, 224, 224) (0, 0, 0) (1, 1, 1) / True Model Inference Error
tresnet_xl (3, 224, 224) (0, 0, 0) (1, 1, 1) / True Model Inference Error
tresnet_xl_448 (3, 448, 448) (0, 0, 0) (1, 1, 1) / True Model Inference Error
tv_densenet121 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 True True
tv_resnet34 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 512 True True
tv_resnet50 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
tv_resnet101 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
tv_resnet152 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
tv_resnext50_32x4d (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
vgg11 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vgg11_bn (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vgg13 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vgg13_bn (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vgg16 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vgg16_bn (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vgg19 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vgg19_bn (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 4096 True True
vit_base_patch16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_patch16_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_patch16_224_miil (3, 224, 224) (0, 0, 0) (1, 1, 1) 768 True True
vit_base_patch16_224_miil_in21k (3, 224, 224) (0, 0, 0) (1, 1, 1) 768 True True
vit_base_patch16_384 (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_patch32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 False True
vit_base_patch32_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_patch32_384 (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_r20_s16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 False True
vit_base_r26_s32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 False True
vit_base_r50_s16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 False True
vit_base_r50_s16_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_r50_s16_384 (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_resnet26d_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 False True
vit_base_resnet50_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_resnet50_384 (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 True True
vit_base_resnet50d_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 False True
vit_deit_base_distilled_patch16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
vit_deit_base_distilled_patch16_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
vit_deit_base_patch16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True True
vit_deit_base_patch16_384 (3, 384, 384) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True True
vit_deit_small_distilled_patch16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
vit_deit_small_patch16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 384 True True
vit_deit_tiny_distilled_patch16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) / True Model Inference Error
vit_deit_tiny_patch16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 192 True True
vit_huge_patch14_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1280 False MMAction2 Test Error
vit_large_patch16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True True
vit_large_patch16_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True True
vit_large_patch16_384 (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True MMAction2 Test Error
vit_large_patch32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 False True
vit_large_patch32_224_in21k (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True True
vit_large_patch32_384 (3, 384, 384) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 1024 True True
vit_large_r50_s32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 768 False True
vit_small_patch16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 True True
vit_small_r20_s16_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 384 False True
vit_small_r20_s16_p2_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 384 False True
vit_small_r26_s32_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 384 False True
vit_small_r_s16_p8_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 384 False True
vit_small_resnet26d_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 False True
vit_small_resnet50d_s16_224 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 768 False True
vit_tiny_r_s16_p8_224 (3, 224, 224) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 192 False True
vovnet39a (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
vovnet57a (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 1024 False True
wide_resnet50_2 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
wide_resnet101_2 (3, 224, 224) (0.485, 0.456, 0.406) (0.229, 0.224, 0.225) 2048 True True
xception (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
xception41 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
xception65 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True
xception71 (3, 299, 299) (0.5, 0.5, 0.5) (0.5, 0.5, 0.5) 2048 True True

TODO

  • test more timm models
  • train a timm based tsn model with kinetics400.
  • changelog

@irvingzhang0512
Copy link
Contributor Author

irvingzhang0512 commented May 22, 2021 •
edited
Loading

image

@kennymckormick typo, right? should be densenet161?
And what does 8x2 of the gpus column mean?

@codecov
Copy link

codecov bot commented May 22, 2021 •
edited
Loading

Codecov Report

Merging #880 (c6f57e9) into master (f8b595a) will decrease coverage by 0.08%.
The diff coverage is 78.57%.

❗ Current head c6f57e9 differs from pull request most recent head fd5531d. Consider uploading reports for the commit fd5531d to get more accurate results
Impacted file tree graph

@@            Coverage Diff             @@
##           master     #880      +/-   ##
==========================================
- Coverage   83.57%   83.49%   -0.09%     
==========================================
  Files         132      132              
  Lines        9966     9977      +11     
  Branches     1718     1720       +2     
==========================================
+ Hits         8329     8330       +1     
- Misses       1218     1226       +8     
- Partials      419      421       +2
Flag Coverage Δ
unittests 83.49% <78.57%> (-0.09%) ⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files Coverage Δ
mmaction/models/recognizers/base.py 68.00% <75.00%> (+1.09%) ⬆️
mmaction/models/recognizers/recognizer2d.py 84.94% <100.00%> (ø)
mmaction/core/evaluation/accuracy.py 92.27% <0.00%> (-0.91%) ⬇️
mmaction/datasets/pipelines/augmentations.py 91.95% <0.00%> (-0.69%) ⬇️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f8b595a...fd5531d. Read the comment docs.

@kennymckormick
Copy link
Member

image

@kennymckormick typo, right? should be densenet161?
And what does 8x2 of the gpus column mean?

Yes, that's a typo. Would you please help correct it in this PR?
Besides, 8x2 means we use 2 nodes for training, each node with 8 gpus (16 in total).

@irvingzhang0512
Copy link
Contributor Author

Thanks for the clarification, will fix the typo

kennymckormick
kennymckormick reviewed May 24, 2021
View reviewed changes
Copy link
Member

@kennymckormick kennymckormick left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice work, please ping me if checkpoints are ready.

@irvingzhang0512
Copy link
Contributor Author

@kennymckormick

tsn-swin-transformer ckpt/log/json could be found here. Results are quite good: top1/5 accuracy are 77.31/92.88

However, since our kinetics400 dataset is quite different(Now I'm using kinetics400 provided by cvdf), I did some tests with tsn-densenet161

Model Accuracy
MMAction2 Model Zoo 72.78/90.75
MMAction2 Model Zoo ckpt tested on my kinetics400 val set 72.01/90.21
Trained/tested on my kinetics400 72.24/90.31

So the accuracy of tsn-swin-transformer ckpt may be different between our datasets. Could you please test the tsn-swin-transformer on your kinetics400 and update accuracy in the TSN readme? If the accuracy is not good enough, maybe you have to help train this model...

@kennymckormick
Copy link
Member

Great, I'm gonna test it on our validation dataset. Besides, I'm going to upload our Kinetics-400 validation dataset.

@kennymckormick
Copy link
Member

Seems some updates need to be committed to pass CI.

@irvingzhang0512
Copy link
Contributor Author

CI raises ImportError, mmcv-full is not properly installed, not sure why that happens

@irvingzhang0512 irvingzhang0512 mentioned this pull request Jun 10, 2021
Closed
@kennymckormick kennymckormick merged commit bf74371 into open-mmlab:master Jun 10, 2021
@irvingzhang0512 irvingzhang0512 deleted the tsn-timm branch June 11, 2021 17:36
@wenjun90 wenjun90 mentioned this pull request Nov 18, 2021
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@kennymckormick kennymckormick kennymckormick approved these changes

@dreamerlin dreamerlin Awaiting requested review from dreamerlin

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@irvingzhang0512 @kennymckormick