/project-next-frame-prediction

Primary LanguagePython

Self-Supervised Model for Video Prediction and Unsupervised Action Recognition

Task Scripts:

  • train_video_prediction.py - Trains a model for the task of video prediction on the UCF101
  • build_classification_dataset.py - Runs videos through the trained video prediction model and stores the intermediate tensors to be used for classifying actions
  • train_video_classification.py - Use the saved intermediate tensors and classify actions.
  • generate_gif.py - Iterates through a few videos of the dataset and generates 6 image frames :- 3 seed frames and 3 predicted frames.
  • tune_hyperparams.py - Hyperparameter tuning using Ray Tune

Configuration:

  • resources/config/user-{linux_or_windows_user}.json - create a similar configuration file specifying the UCF101 dataset location, number of workers for processing, classification dataset directory, etc

Instructions to run script:

  1. Make sure you have created a user config.

  2. Please set the PYTHONPATH environment variable in order to run the scripts.

  3. You can execute each script by calling python script_file.py

Sample Output

YOYO