This is a pytorch implementation of Dense Events Propagation Network (DepNet) on ActivityNet Captions for the AAAI 2021 oral paper "Dense Events Grounding in Video" .
Please download the visual features from the official website of ActivityNet: Official C3D Feature. And you can download preprocessed annotation files here.
- python 3.5
- pytorch 1.4.0
- torchtext
- easydict
- terminaltables
Use the following commands for training:
cd moment_localization && export CUDA_VISIBLE_DEVICES=0
python dense_train.py --verbose --cfg ../experiments/dense_activitynet/acnet.yaml
You may get better results than that reported in our paper thanks to the code updates.
If you use our code or models in your research, please cite with:
@inproceedings{bao2021dense,
title = {Dense Events Grounding in Video},
author = {Bao, Peijun and Zheng, Qian and Mu, Yadong},
booktitle = {AAAI},
year = {2021}
}