Tag
DVD-JEPA is an open-source, minimal JEPA world model that learns representations from video by predicting future embeddings rather than pixels. It uses a bouncing DVD logo to demonstrate position recovery, dreaming, and anomaly detection, all running in a browser.