MatAnyone-E: Real-Time 4K Video Matting

Loading video
Loading videoMatAnyone-E uses a low-resolution pass to find hard regions such as hair, then runs sparse high-resolution matching and a lightweight decoder; at 4K it reports a 7.4x speedup, 7.6% of prior peak GPU memory, and 28.6 FPS on one H200.
Video MattingHigh-Resolution VideoSparse Memory PropagationLightweight DecoderMatAnyone-ENanyang Technological UniversityS-LabVideo Processing
Category: video-generation
Author: @Gorden_Sun
Date: 2026-11-08T00:00:00
Duration: 58.176s





