vsumm-reinforce

This is the official implementation of the AAAI'18 paper Deep Reinforcement Learning for Unsupervised Video Summarization with Diversity-Representativeness Reward. The code is based on Theano (version 0.9.0).

Pytorch implementation can be found here.

Preparation

To get the datasets and models, you will need wget.

Run the following commands in order

git clone https://github.com/KaiyangZhou/vsumm-reinforce
cd vsumm-reinforce
# download datasets.tar.gz
wget http://www.eecs.qmul.ac.uk/~kz303/vsumm-reinforce/datasets.tar.gz
tar -xvzf datasets.tar.gz
# download models.tar.gz
wget http://www.eecs.qmul.ac.uk/~kz303/vsumm-reinforce/models.tar.gz
tar -xvzf models.tar.gz

Updates: The QMUL server is inaccessible. Download the datasets from this google drive link.

How to train

Training code is implemented in vsum_train.py. To train a RNN, run

python vsum_train.py --dataset datasets/eccv16_dataset_tvsum_google_pool5.h5 --max-epochs 60 --hidden-dim 256

How to test

Test code is implemented in vsum_test.py. For example, to test with our models, simply run

python vsum_test.py -model models/model_tvsum_reinforceRNN.h5 -d tvsum
python vsum_test.py -model models/model_tvsum_reinforceRNN_sup.h5 -d tvsum
python vsum_test.py -model models/model_summe_reinforceRNN.h5 -d summe
python vsum_test.py -model models/model_summe_reinforceRNN_sup.h5 -d summe

Output results are saved to log-test/results.h5. To visualize score-vs-gtscore, you can use visualize_results.py by

python visualize_results.py -p log-test/result.h5

Visualize summary

You can use summary2video.py to transform the binary machine_summary to real summary video. You need to have a directory containing video frames. The code will automatically write summary frames to a video where the frame rate can be controlled. Use the following command to generate a .mp4 video

python summary2video.py -p path_to/result.h5 -d path_to/video_frames -i 0 --fps 30 --save-dir log --save-name summary.mp4

Please remember to specify the naming format of your video frames on this line.

Citation

@article{zhou2017reinforcevsumm, 
   title={Deep Reinforcement Learning for Unsupervised Video Summarization with Diversity-Representativeness Reward},
   author={Zhou, Kaiyang and Qiao, Yu and Xiang, Tao}, 
   journal={arXiv:1801.00054}, 
   year={2017} 
}