Cross-Architecture Self-supervised Video Representation Learning arxiv

Usage

Requirements

• python3.6
• ffmpeg-3.3.30-4
• tensorboard==1.15
• opencv-python==4.2.0.34
• python-Levenshtein==0.12.0
• scikit-video==1.1.11
• torchvision-0.5.0
• torch-1.4.0

Pretrain SDP

python train_pretrain_sdp.py --cl 16 --it 2\
    --log pretrain_logs/sdp_c3d_ucf101 --model c3d\
    --dist-url 'tcp://localhost:10001' --multiprocessing-distributed\
    --world-size 1 --rank 0 --bs 64\
    --pf 10 --aug-plus --cos --epochs 300\
    --workers 32

Pretrain CACL

python train_pretrain.py --cl 16 --it 2\
    --log pretrain_logs/cacl_c3d_ucf101 --model c3d\
    --dist-url 'tcp://localhost:10001' --multiprocessing-distributed\
    --world-size 1 --rank 0 --bs 64\
    --pf 10 --aug-plus --cos --epochs 300\
    --workers 32

Retrieve videos

python retrieve_videos.py --feature_dir [save_feature_dir] --bs 8 --ckpt [pretrained_weight.pth] --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0 --workers 8

Fintune

python train_finetune.py --dist-url 'tcp://localhost:10001' \
    --multiprocessing-distributed --world-size 1 \
    --rank 0 --bs 64 --pf 10 --epochs 150 \
    --workers 8 --log finetune_log/cacl_ucf_c3d --lr 0.1 --ckpt [pretrained_weight.pth] --cos

If you find our code or paper useful, please cite as

@InProceedings{guo_2022_CVPR,
    author    = {Guo, Sheng and Xiong, Zihua and Zhong, Yujie and Wang, Limin Wang and Guo, Xiaobo and Han, Bing and Huang Weilin},
    title     = {Cross-Architecture Self-supervised Video Representation Learning},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2022}
}