This project focuses on extreme video compression using pre-trained diffusion models in Python 3.8 with Conda, including GPU usage. The input for the project is a 5-dimensional array with specific frame sizes. An example array from the Cityscape dataset is provided for reference. To measure compression metrics, code for H.264 and H.265 is available. Checkpoints for video generation and image compression are also included. The model’s performance exceeds traditional standards like H.264 and H.265 at low bitrates. This is showcased using data from videos in the city of Bonn. Unique features include the utilization of diffusion models for compression and the comparison with established standards.
https://github.com/ElesionKyrie/Extreme-Video-Compression-With-Prediction-Using-Pre-trainded-Diffusion-Models-