This repo holds the codes and models for the DRN framework presented on CVPR 2020

Dense Regression Network for Video Grounding Runhao Zeng, Haoming Xu, Wenbing Huang, Peihao Chen, Mingkui Tan, Chuang Gan, CVPR 2020, Seattle, Washington.

[Paper]

• Usage Guide
  • Module Preparation
  • Code Preparation
  • Training DRN
    • First Stage
    • Second Stage
    • Third Stage
  • Testing DRN
• Other Info
  • Citation
  • Contact

[back to top]

Clone this repo with git

git clone https://github.com/Alvin-Zeng/DRN
cd DRN

Here, we provide the C3D features on Charades-STA for training and testing.

Charades-STA: You can download it from Baidu Cloud (password: smil).

[back to top]

Start from a clear conda env

conda create -n DRN
conda activate DRN

This repo is based on FCOS, use the following command to install it

bash setup.sh

Other minor Python modules can be installed by running

pip install -r requirements.txt

[back to top]

Plesse first set the path of features in data/default_config.yaml

feature_root: $PATH_OF_FEATURES

Use the following command to train the first stage of DRN

bash drn_train.sh $PATH_TO_SAVE_FIRST_MODEL is_first_stage

• $PATH_TO_SAVE_FIRST_MODEL denotes the path to save the first-stage model

Use the following command to train the second stage of DRN

bash drn_train.sh $PATH_TO_SAVE_SECOND_MODEL is_second_stage $FIRST_CHECKPOINT 

• $PATH_TO_SAVE_SECOND_MODEL denotes the path to save the second-stage model

• $FIRST_CHECKPOINT denotes the trained model from the first stage

Use the following command to train the third stage of DRN

bash drn_train.sh $PATH_TO_SAVE_THIRD_MODEL is_third_stage $SECOND_CHECKPOINT

• $PATH_TO_SAVE_THIRD_MODEL denotes the path to save the third-stage model

• $SECOND_CHECKPOINT denotes the trained model from the second stage

[back to top]

Here, we provide the models trained on Charades-STA for testing.

Charades-STA: You can download them from Baidu Cloud (password: smil).

Use the following command to test the trained model

bash drn_test.sh $TRAINED_CHECKPOINT

• $TRAINED_CHECKPOINT denotes the trained model

The evaluation results will be put in the "results" folder

[back to top]

Please cite the following paper if you feel DRN useful to your research

@inproceedings{DRN2020CVPR,
  author    = {Runhao Zeng and
               Haoming Xu and
               Wenbing Huang and
               Peihao Chen and
               Mingkui Tan and
               Chuang Gan},
  title     = {Dense Regression Network for Video Grounding},
  booktitle = {CVPR},
  year      = {2020},
}

For any question, please file an issue or contact

Runhao Zeng: [email protected]