Combining deep features and activity context via conditional random fields to improve activity recognition

1. To prepare data, run prepare_all.py
2. To learn and infer with the model, run predict.py
3. The dataset for training and testing the C3D model can be found at https://drive.google.com/open?id=1mEnLUrnElfbqmfmQWPaBK8V-JX8zZ_sV
4. To visualize the results, run viz.py