This repo is the build for the 2nd task of VidVRD: Video Relation Prediction
The 2st stage project: Video Action Detection
To use this repository, you first need to extract the Vidor video segment features following the other repo Vidor-i3d
train_tgm.py contains the updated codes for training the tgm now compatiable with python 3.
E.g.
python train_tgm.py
vidor_i3d_per_video.py contains the codes for loading extracted features from Vidor Dataset