This is our implementation for the paper:
Ning Han, Jingjing Chen, Chuhao Shi, Yawen Zeng, Guangyi Xiao, and Hao Chen. 2022. BiC-Net: Learning Efficient Spatio-Temporal Relation for Text-Video Retrieval.
We use the framework pytorch.
You can also follow the instruction below to train your own model.
Run train.py to train and save models:
python train.py --cuda --is_train --dataset=msr-vtt --data_split=9000 --layer_num=4 --log_dir=./data/runs/xxx --dataroot=./data/MSR-VTT
run eval.py to evaluate models:
python eval.py --cuda --checkpoint= ./models/ckpt_best.pth
There are a lot of experimental records in the ./data/runs/xxx
We provide three datasets that we used in our paper: MSR-VTT, MSVD, YouCook2.
Download the processed video and text features of MSR-VTT(code:pbvc), MSVD(code:5p0y), and YouCook2_BB, and save them in /data
folder.
Last Update Date: May 29, 2022