[CodeCamp2023-543] Adapt new version of Config

configs/_base_/models/tin_r50.py

@@ -1,7 +1,9 @@
 # model settings
 
 preprocess_cfg = dict(
-    mean=[127.5, 127.5, 127.5], std=[127.5, 127.5, 127.5], format_shape='NCHW')
+    mean=[123.675, 116.28, 103.53],
+    std=[58.395, 57.12, 57.375],
+    format_shape='NCHW')
 
 model = dict(
     type='Recognizer2D',

configs/recognition/tin/README.md

@@ -34,7 +34,7 @@ For a long time, the vision community tries to learn the spatio-temporal represe
 
 | frame sampling strategy |   resolution   | gpus | backbone |    pretrain     | top1 acc | top5 acc | testing protocol | inference time(video/s) | gpu_mem(M) |          config           |          ckpt           |           log           |
 | :---------------------: | :------------: | :--: | :------: | :-------------: | :------: | :------: | :--------------: | :---------------------: | :--------: | :-----------------------: | :---------------------: | :---------------------: |
-|          1x1x8          | short-side 256 | 8x4  | ResNet50 | TSM-Kinetics400 |  71.77   |  90.36   | 8 clips x 1 crop |            x            |    6185    | [config](/configs/recognition/tin/tin_imagenet-pretrained-r50_8xb6-1x1x8-40e_sthv2-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/tin/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb_20220913-7f10d0c0.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/tin/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb.log) |
+|          1x1x8          | short-side 256 | 8x4  | ResNet50 | TSM-Kinetics400 |  71.86   |  90.44   | 8 clips x 1 crop |            x            |    6185    | [config](/configs/recognition/tin/tin_imagenet-pretrained-r50_8xb6-1x1x8-40e_sthv2-rgb.py) | [ckpt](https://download.openmmlab.com/mmaction/v1.0/recognition/tin/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb_20220913-7f10d0c0.pth) | [log](https://download.openmmlab.com/mmaction/v1.0/recognition/tin/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb.log) |
 
 Here, we use `finetune` to indicate that we use [TSM model](https://download.openmmlab.com/mmaction/v1.0/v1.0/recognition/tsm/tsm_imagenet-pretrained-r50_8xb16-1x1x8-50e_kinetics400-rgb/tsm_imagenet-pretrained-r50_8xb16-1x1x8-50e_kinetics400-rgb_20220831-64d69186.pth) trained on Kinetics-400 to finetune the TIN model on Kinetics-400.
 

configs/recognition/tin/metafile.yml

@@ -66,8 +66,8 @@ Models:
   Results:
   - Dataset: Kinetics-400
     Metrics:
-      Top 1 Accuracy: 71.77
-      Top 5 Accuracy: 90.36
+      Top 1 Accuracy: 71.86
+      Top 5 Accuracy: 90.44
     Task: Action Recognition
   Training Log: https://download.openmmlab.com/mmaction/v1.0/recognition/tin/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb.log
   Weights: https://download.openmmlab.com/mmaction/v1.0/recognition/tin/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb/tin_kinetics400-pretrained-tsm-r50_1x1x8-50e_kinetics400-rgb_20220913-7f10d0c0.pth

dataset-index.yml

@@ -1,39 +1,40 @@
+openxlab: true
 kinetics400:
-  dataset: Kinetics-400
+  dataset: OpenMMLab/Kinetics-400
   download_root: data
   data_root: data/kinetics400
-  script: tools/data/kinetics/k400_preprocess.sh
+  script: tools/data/kinetics/preprocess_k400.sh
 
 kinetics600:
-  dataset: Kinetics600
+  dataset: OpenMMLab/Kinetics600
   download_root: data
   data_root: data/kinetics600
-  script: tools/data/kinetics/k600_preprocess.sh
+  script: tools/data/kinetics/preprocess_k600.sh
 
 kinetics700:
-  dataset: Kinetics_700
+  dataset: OpenMMLab/Kinetics_700
   download_root: data
   data_root: data/kinetics700
-  script: tools/data/kinetics/k700_preprocess.sh
+  script: tools/data/kinetics/preprocess_k700.sh
 
 sthv2:
-  dataset: sthv2
+  dataset: OpenDataLab/sthv2
   download_root: data
   data_root: data/sthv2
   script: tools/data/sthv2/preprocess.sh
 
 ucf-101:
-  dataset: UCF101
+  dataset: OpenDataLab/UCF101
   download_root: data
   data_root: data/ucf101
 
 finegym:
-  dataset: FineGym
+  dataset: OpenDataLab/FineGym
   download_root: data
   data_root: data/gym
 
 diving48:
-  dataset: diving48
+  dataset: OpenDataLab/diving48
   download_root: data
   data_root: data/diving48
   script: tools/data/diving48/preprocess.sh

mmaction/configs/recognition/uniformerv2/uniformerv2-base-p16-res224_clip-kinetics710-kinetics-k400-pre_16xb32-u8_mitv1-rgb.py

@@ -0,0 +1,178 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+from mmengine.config import read_base
+
+with read_base():
+    from ..._base_.default_runtime import *
+
+from mmengine.dataset import DefaultSampler
+from mmengine.optim import CosineAnnealingLR, LinearLR
+from mmengine.runner import EpochBasedTrainLoop, TestLoop, ValLoop
+from torch.optim import AdamW
+
+from mmaction.datasets import (CenterCrop, DecordDecode, DecordInit, Flip,
+                               FormatShape, PackActionInputs,
+                               PytorchVideoWrapper, RandomResizedCrop, Resize,
+                               ThreeCrop, UniformSample, VideoDataset)
+from mmaction.evaluation import AccMetric
+from mmaction.models import (ActionDataPreprocessor, Recognizer3D,
+                             TimeSformerHead, UniFormerHead, UniFormerV2)
+
+# model settings
+num_frames = 8
+model = dict(
+    type=Recognizer3D,
+    backbone=dict(
+        type=UniFormerV2,
+        input_resolution=224,
+        patch_size=16,
+        width=768,
+        layers=12,
+        heads=12,
+        t_size=num_frames,
+        dw_reduction=1.5,
+        backbone_drop_path_rate=0.,
+        temporal_downsample=False,
+        no_lmhra=True,
+        double_lmhra=True,
+        return_list=[8, 9, 10, 11],
+        n_layers=4,
+        n_dim=768,
+        n_head=12,
+        mlp_factor=4.,
+        drop_path_rate=0.,
+        mlp_dropout=[0.5, 0.5, 0.5, 0.5],
+        clip_pretrained=False,
+        init_cfg=dict(
+            type='Pretrained',
+            checkpoint=  # noqa: E251
+            'https://download.openmmlab.com/mmaction/v1.0/recognition/uniformerv2/kinetics400/uniformerv2-base-p16-res224_clip-kinetics710-pre_u8_kinetics400-rgb_20221219-203d6aac.pth',  # noqa: E501
+            prefix='backbone.')),
+    cls_head=dict(
+        type=TimeSformerHead,
+        dropout_ratio=0.5,
+        num_classes=339,
+        in_channels=768,
+        average_clips='prob'),
+    data_preprocessor=dict(
+        type=ActionDataPreprocessor,
+        mean=[114.75, 114.75, 114.75],
+        std=[57.375, 57.375, 57.375],
+        format_shape='NCTHW'))
+
+# dataset settings
+dataset_type = 'VideoDataset'
+data_root = 'data/mit/videos/training'
+data_root_val = 'data/mit/videos/validation'
+ann_file_train = 'data/mit/mit_train_list_videos.txt'
+ann_file_val = 'data/mit/mit_val_list_videos.txt'
+ann_file_test = 'data/mit/mit_val_list_videos.txt'
+
+file_client_args = dict(io_backend='disk')
+train_pipeline = [
+    dict(type=DecordInit, **file_client_args),
+    dict(type=UniformSample, clip_len=num_frames, num_clips=1),
+    dict(type=DecordDecode),
+    dict(type=Resize, scale=(-1, 256)),
+    dict(
+        type=PytorchVideoWrapper, op='RandAugment', magnitude=7, num_layers=4),
+    dict(type=RandomResizedCrop),
+    dict(type=Resize, scale=(224, 224), keep_ratio=False),
+    dict(type=Flip, flip_ratio=0.5),
+    dict(type=FormatShape, input_format='NCTHW'),
+    dict(type=PackActionInputs)
+]
+
+val_pipeline = [
+    dict(type=DecordInit, **file_client_args),
+    dict(type=UniformSample, clip_len=num_frames, num_clips=1, test_mode=True),
+    dict(type=DecordDecode),
+    dict(type=Resize, scale=(-1, 224)),
+    dict(type=CenterCrop, crop_size=224),
+    dict(type=FormatShape, input_format='NCTHW'),
+    dict(type=PackActionInputs)
+]
+
+test_pipeline = [
+    dict(type=DecordInit, **file_client_args),
+    dict(type=UniformSample, clip_len=num_frames, num_clips=4, test_mode=True),
+    dict(type=DecordDecode),
+    dict(type=Resize, scale=(-1, 224)),
+    dict(type=ThreeCrop, crop_size=224),
+    dict(type=FormatShape, input_format='NCTHW'),
+    dict(type=PackActionInputs)
+]
+
+train_dataloader = dict(
+    batch_size=8,
+    num_workers=8,
+    persistent_workers=True,
+    sampler=dict(type=DefaultSampler, shuffle=True),
+    dataset=dict(
+        type=VideoDataset,
+        ann_file=ann_file_train,
+        data_prefix=dict(video=data_root),
+        pipeline=train_pipeline))
+val_dataloader = dict(
+    batch_size=8,
+    num_workers=8,
+    persistent_workers=True,
+    sampler=dict(type=DefaultSampler, shuffle=False),
+    dataset=dict(
+        type=VideoDataset,
+        ann_file=ann_file_val,
+        data_prefix=dict(video=data_root_val),
+        pipeline=val_pipeline,
+        test_mode=True))
+test_dataloader = dict(
+    batch_size=8,
+    num_workers=8,
+    persistent_workers=True,
+    sampler=dict(type=DefaultSampler, shuffle=False),
+    dataset=dict(
+        type=VideoDataset,
+        ann_file=ann_file_test,
+        data_prefix=dict(video=data_root_val),
+        pipeline=test_pipeline,
+        test_mode=True))
+
+val_evaluator = dict(type=AccMetric)
+test_evaluator = dict(type=AccMetric)
+train_cfg = dict(
+    type=EpochBasedTrainLoop, max_epochs=24, val_begin=1, val_interval=1)
+val_cfg = dict(type=ValLoop)
+test_cfg = dict(type=TestLoop)
+
+base_lr = 2e-5
+optim_wrapper = dict(
+    optimizer=dict(
+        type=AdamW, lr=base_lr, betas=(0.9, 0.999), weight_decay=0.05),
+    paramwise_cfg=dict(norm_decay_mult=0.0, bias_decay_mult=0.0),
+    clip_grad=dict(max_norm=20, norm_type=2))
+
+param_scheduler = [
+    dict(
+        type=LinearLR,
+        start_factor=1 / 20,
+        by_epoch=True,
+        begin=0,
+        end=5,
+        convert_to_iter_based=True),
+    dict(
+        type=CosineAnnealingLR,
+        eta_min_ratio=1 / 20,
+        by_epoch=True,
+        begin=5,
+        end=24,
+        convert_to_iter_based=True)
+]
+
+default_hooks.update(
+    dict(
+        checkpoint=dict(interval=3, max_keep_ckpts=5),
+        logger=dict(interval=100)))
+
+# Default setting for scaling LR automatically
+#   - `enable` means enable scaling LR automatically
+#       or not by default.
+#   - `base_batch_size` = (8 GPUs) x (8 samples per GPU).
+auto_scale_lr = dict(enable=True, base_batch_size=512)