A PyTorch implementation of APN, a differentiable framework named adversarial pyramid network is a new deep architecture thatimproves the generalizability of video features by capturing the local-relation, global-relation, and cross-relation features progressively. Based on these pyramid features, we also propose a new adversarial data augmentation approach that can enhance the diversity and qualityof the augmented data to bridge different video domains.

This paper is accepted by TPAMI2021.

1. Install Python 3.7, PyTorch 1.3, and OpenCV 3.4.

2. Train the model. You can use the following bash script to train the model.

sh train.sh

You may send email to yaozy19@mails.tsinghua.edu.cn, yunbo.thu@gmail.com or longmingsheng@gmail.com.