README.md

Spatiotemporal-Residual-Propagation

Code release for ICCV 2019 paper "Spatiotemporal Feature Residual Propagation for Action Prediction;"

Environment Requirements:

1. Python 2.7
2. Pytorch 0.4.0
3. CUDA-8
4. ffmpeg
5. PIL

Dataset Preparation

Download

• JHMDB: http://jhmdb.is.tue.mpg.de/challenge/JHMDB/datasets
• BIT: https://sites.google.com/site/alexkongy/software
• UCF101: https://www.crcv.ucf.edu/data/UCF101.php

For dataset loading and augmenting, we modify code from tsn (https://github.com/yjxiong/tsn-pytorch/blob/master/dataset.py) for every dataset and the modified version are locating in each sub-folder.

Pretraining Process:

Since proposed method focus on feature propagation, we obtain classifiers offline by pretraining each dataset with TSN model.

• UCF-101: Directly from tsn pretrain;
• JHMDB-21: Pretrain on RGB;
• BIT: Pretrain on Flow;

See pretrain folder in each subfolder for more details; Pretrain Model Link:

• JHMDB-21 Split 1: https://drive.google.com/open?id=1dc8Ec6FAg5N-gHfW65f_mXpP4mlDzlpv;
• BIT: https://drive.google.com/file/d/14uZRSebu3pNraRjZfnmKl41jJeTDaSNM/view?usp=sharing
• UCF-101 Split 1 flow: https://drive.google.com/open?id=1o8A8-4OKVlJL-QmOmSfppcUTyWxrMX1Q;
• UCF-101 Split 1 rgb: https://drive.google.com/open?id=1gXUrIBaf98OrVKHUy-dn2khC2EH3zK8X;

Training and Comparison

Please see details in each subfolder for training and testing procedures