Real-Time Intermediate Flow Estimation (RIFE)

RIFE (Real-Time Intermediate Flow Estimation) is an open-source project implementing video frame interpolation using deep learning. It estimates intermediate frames between existing video frames to increase the frame rate or create slow-motion effects. The core architecture involves optical flow estimation and refinement networks, allowing the generation of new frames based on the movement and appearance information extracted from the input frames. It can run at 30+ FPS for 2x 720p interpolation on a 2080Ti GPU. RIFE supports arbitrary-timestep interpolation. The model is implemented in Python using PyTorch, and Dockerfiles are provided for containerization and easy deployment. It also features pre-trained models for HD video and supports GPU acceleration. It has been optimized for anime scenes and diffusion model generated videos. It enables functionalities like video stitching and optical flow estimation.