This repository contains the code for building a system similar to the one from the work Video Description using Bidirectional Recurrent Neural Networks, presented at the International Conference of Artificial Neural Networks (ICANN'16). With this module, you can replicate our experiments and easily deploy new models. ABiViRNet is built upon our fork of Keras framework and tested for the Theano and Tensorflow backends.

• Attention model over the input sequence of frames
• Peeked decoder LSTM: The previously generated word is an input of the current LSTM timestep
• MLPs for initializing the LSTM hidden and memory state
• Beam search decoding

ABiViRNet requires the following libraries:

• Our version of Keras v1.2.1
• Multimodal Keras Wrapper v0.7
• Coco-caption evaluation package

Assuming you have a dataset and features extracted from the video frames:

1. Prepare data:

python data_engine/subsample_frames_features.py

python data_engine/generate_features_lists.py

python data_engine/generate_descriptions_lists.py

See data_engine/README.md for detailed information.

2. Prepare the inputs/outputs of your model in data_engine/prepare_data.py

3. Set a model configuration in config.py

4. Train!:

python main.py

If you use this code for any purpose, please, do not forget to cite the following paper:

Peris, Á., Bolanos, M., Radeva, P., & Casacuberta, F. (2016, September). Video description using bidirectional recurrent neural networks. In International Conference on Artificial Neural Networks (pp. 3-11). Springer International Publishing.

Bibtex version:

@inproceedings{peris2016video,
  title={Video description using bidirectional recurrent neural networks},
  author={Peris, {\'A}lvaro and Bolanos, Marc and Radeva, Petia and Casacuberta, Francisco},
  booktitle={International Conference on Artificial Neural Networks},
  pages={3--11},
  year={2016},
  organization={Springer}
}

Joint collaboration between the Computer Vision at the University of Barcelona (CVUB) group at Universitat de Barcelona-CVC and the PRHLT Research Center at Universitat Politècnica de València.

Álvaro Peris (web page): [email protected]

Marc Bolaños (web page): [email protected]