rmdn_comparison

experiments/rmdn_comparison

Some code to train and predict dreyeve sequences out of the Recurrent Mixture Density Network.

• batch_generators.py holds code to generate training batches.
• callbacks.py holds Keras callbacks called during training.
• compute_c3d_features.py is a script that computes C3D encodings for a complete dreyeve sequence.

Usage: python compute_c3d_features.py --seq <seq>

• config.py holds some shared configuration parameters.
• models.py holds two keras models, namely the C3D encoder to compute features and the RMDN model.
• objectives.py holds the Keras backend code for the negative log likelyhood loss.
• predict_dreyeve_sequence.py is a script that can be used to compute predictions all over a sequence.

Usage: python retrieve_c3d_encodings_from_cineca.py --host <host> --user <user> --password <password>

• retrieve_c3d_encodings_from_cineca.py is a script that retrieves all c3d precomputed encodings from cineca server with sftp.
• train.py is the main training script.
• utils.py holds utility functions.