This is the implementation of ''Spatial-temporal Causal Inference for Partial Image-to-video Adaptation (AAAI2021)'' using Pytorch.
-
source data list (put it in opt.dataroot/list/sourcelistname.txt)
The format of sourcelistname.txt is the resnet50 feature of an image and the corresponding label- For example:
home/chenjin/HPDA/data/Stanford_resnet50_feature/WritingOnBoard/writing_on_a_board_143.npy 11 home/chenjin/HPDA/data/Stanford_resnet50_feature/WritingOnBoard/writing_on_a_board_170.npy 11 ...
-
target data list (put it in opt.dataroot/list/targetlistname.txt)
The format of targetlistname.txt is the i3d feature of a video, the corresponding label, the number of frames of this video- For example:
home/chenjin/HPDA/data/ucf_i3d_feature/WritingOnBoard/v_WritingOnBoard_g11_c05.npy 11 74 home/chenjin/HPDA/data/ucf_i3d_feature/WritingOnBoard/v_WritingOnBoard_g19_c01.npy 11 74 ...
-
Prepare video frame features (saved in dataroot/opt.F_path)
The format of video frame features is ``opt.F_path/class_dir/video_name/framenum.npy'' -
Features are avaiable in https://pan.baidu.com/s/11DlH06gz2cjRPOoOnB-KEw with token: xf2r
- For S-U task, bash run_st_concat_pseudo_SU.sh
- For E-H task, bash run_st_concat_pseudo_EH.sh
@inproceedings{chen2021spatial,
title={Spatial-temporal Causal Inference for Partial Image-to-video Adaptation},
author={Chen, Jin and Wu, Xinxiao and Yao Hu and Jiebo, Luo},
booktitle={Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)},
year={2021}
}