MatAnyone-E brings 7.4x faster 4K video matting via sparse memory propagation
mmlab-ntu · hf · 2026-10-28
Researchers from The University of Queensland and NTU's S-Lab introduce MatAnyone-E for efficient high-resolution video matting.
- Sparse memory propagation: high-resolution matching is applied selectively to uncertain regions, with a lightweight fusion decoder handling the rest
- 3D sparse memory propagation plus a lightweight decoder cuts runtime (7.4x faster) and peak GPU memory (13.7x lower) per the paper's figures
- Preserves matting quality on native 4K (3840×2160) input, with hair-detail results beating VideoMaMa and SAM2Matting baselines
A useful reference for high-resolution matting in post-production and generative workflows.
Related event: MatAnyone-E Speeds Up 4K Video Matting 7.4x With Just 7.6% Memory(4 posts)→
More from Multimodal
- Seedance 2.5 hailed as taking AI filmmaking to another level — SimplyAnnisa · 2026-10-28
- Hands-on with Baidu's Unlimited OCR 2: whole docs in one pass, no page limit — Baidu_Inc · 2026-10-28
- Agentic video editing goes pro: senior editors 5x output, costs to collapse — dotey · 2026-10-28
- Alibaba's OvisOCR3: a 0.8B model tops OmniDocBench at 96.74 via multi-teacher distillation + RL — ATH-MaaS · 2026-10-28
- CodeCap-Bench: executable code beats natural language as video captions — Jiafu Tang · 2026-10-28
- Moving Atoms Engine turns real-world spaces and cities into high-resolution 4D models for sim — Scobleizer · 2026-10-28