The code is written in pytorch and its base is from https://github.com/xdshang/VidVRD-helper. When I want to reproduce those results of this paper----- Video Visual Relation Detection, some problems emerge. Compared to VidVRD,other baselines in this paper produce worse results but my VTransE's version produce better results.
please read "实验说明.docx" before you use this code and download relative datasets.
python main.py --batch_size 64 ...other parameters