Multiple-Attention-Model-for-MovieQA

Extracting Image Features from Movie Clips

1. Download the MovieQA dataset
2. Install ffmpeg, keras, tensorflow
3. git clone https://github.com/fchollet/deep-learning-models
4. Edit paths in movie2feature.py
5. RunPython movie2feature.py

When I use the setting where videos are sliced every 2 seconds, the final processed data is 73GB. and the overall process took 5 hours to run on a GTX 1080Ti.