Awesome

Video Test-Time Adaptation for Action Recognition (CVPR 2023)

ProjectPage

ViTTA is the first approach of test-time adaptation of video action recognition models against common distribution shifts. ViTTA is tailored to saptio-temporal models and capable of adaptation on a single video sample at a step. It consists in a feature distribution alignment technique that aligns online estimates of test set statistics towards the training statistics. It further enforces prediction consistency over temporally augmented views of the same test video sample.

Official implementation of ViTTA [arXiv]
Author HomePage
🤗 Dataset (12 corruption types of Kinetics 400 and Something-Something v2, and UCF101 data)

Requirements

• Our experiments run on Python 3.6, PyTorch 1.7, mmcv-full 1.3.12. Other versions should work but are not tested.
• Dependency of mmaction2 (for Video Swin Transformer):
  $ pip install mmcv-full==1.3.12
  $ git clone https://github.com/SwinTransformer/Video-Swin-Transformer.git && cd Video-Swin-Transformer
  $ pip install -v -e . --user
• Other relevant dependencies can be found in requirements.txt

Data Preparation

• Download
  Download required data for Experiments on UCF101 from here
  list_video_perturbations_ucf: list of files for corrupted videos of UCF101 validation set (in 12 corruption types)
  model_swin_ucf: Video Swin Transformer trained on UCF101 training set
  model_tanet_ucf: TANet trained on UCF101 training set
  model_tanet_ucf: TANet trained on UCF101 training set
  source_statistics_tanet_ucf: precomputed source (UCF101 training set) statistics on TANet
  source_statistics_swin_ucf: precomputed source (UCF101 training set) statistics on Video Swin Transformer
  ucf_corrupted_videos: a folder of 12 compressed files containing videos of UCF validation set (in 12 corruption types)
  ucf_corrupted_videos.zip: a single compressed file (83.8GB) containing videos of UCF validation set (in 12 corruption types)
• Data structure
  lines in file list are in format video_path n_frames class_id
  video dataset structure

  level_5_ucf_val_split_1_/
    gauss/
      ApplyEyeMakeup/
        v_ApplyEyeMakeup_g01_c01.mp4
        v_ApplyEyeMakeup_g01_c02.mp4
        ...
      ApplyLipstick/
      ...
    contrast/
      ApplyEyeMakeup/
        v_ApplyEyeMakeup_g01_c01.mp4
        v_ApplyEyeMakeup_g01_c02.mp4
        ...
      ApplyLipstick/
    ...
  

Usage

Specify the data paths in the scripts correspondingly (see comments in scripts)

• Precompute source statistics on training set
  precompute source (UCF101 training set) statistics on TANet:
  $ python compute_stats/compute_spatiotemp_stats_clean_train_tanet.py
  precompute source (UCF101 training set) statistics on Video Swin Transformer:
  $ python compute_stats/compute_spatiotemp_stats_clean_train_swin.py
• Test-time adaptation
  $ python tta_tanet_ucf101.py test-time adaptation on TANet UCF101
  $ python tta_swin_ucf101.py test-time adaptation on Video Swin Transformer UCF101
• Source-only evaluation on corrupted validation data
  $ python tta_tanet_ucf101.py
  $ python tta_swin_ucf101.py

Citation

Thanks for citing our paper:

@inproceedings{lin2023video,
  title={Video Test-Time Adaptation for Action Recognition},
  author={Lin, Wei and Mirza, Muhammad Jehanzeb and Kozinski, Mateusz and Possegger, Horst and Kuehne, Hilde and Bischof, Horst},
  booktitle={CVPR},
  year={2023},
}