Newest papers

VideoMaMa++: Temporally Consistent Video Matting via Preserve-and-Refine Token

Sangbeom Lim, Seoung Wug Oh, Heeji Yoon, Seungryong Kim, Joon-Young Lee · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26857 · cited by 0 works (OpenAlex) · open-access PDF

German

Lange Aufnahmen-basiertes Video-Matting hat sich vor kurzem als führende Herangehensweise entwickelt, indem Video-Frame-Priors von Internet-Skala-Prätraining ausgenutzt werden, um eine starke Null-Shot-Generalisierung aus begrenzten synthetischen Daten zu erzielen. Allerdings begrenzt die schwere Berechnung dieser Diffusion-Rackbones die Inferenz auf ein kurzes Stück von nur 6 bis 24 Frames, so dass selbst gewöhnliche Videos mehrere Stücke umfassen. Weil jeder Stück unabhängig von groben Maskenleitungen mattiert wird, scheitern bestehende Methoden, die Konsistenz über Schunkgrenzen hinweg aufrechtzuerhalten, und produzieren blinkende Kanten und diskontinuierliche Transparenz. Wir behandeln dies mit einem

Machine translation (Meta M2M100); the English original follows.

Abstract

Diffusion-based video matting has recently emerged as a leading approach, leveraging video diffusion priors from internet-scale pretraining to achieve strong zero-shot generalization from limited synthetic data. However, the heavy compute of these diffusion backbones limits inference to a short chunk of just 6 to 24 frames, so even ordinary videos span multiple chunks. Because each chunk is matted independently from coarse mask guidance, existing methods fail to maintain consistency across chunk boundaries, producing flickering edges and discontinuous transparency. We address this with a mask-guided diffusion-matting framework built on two ideas. The RGB and mask conditioning are represented as separate token streams that interact through full spatio-temporal self-attention, letting the model adaptively combine fine RGB texture with coarse mask layout rather than committing to a fixed channel-wise fusion. On top of this, we propose a novel pair of learnable embeddings, the preserve token and the refine token, that act as a per-frame conditioning interface and enable a sliding-window inference scheme in which already-generated mattes are propagated across chunk boundaries for local temporal consistency, complemented by a globally shared reference anchor at every chunk to prevent error accumulation in long sequences without any additional training. Across standard video matting benchmarks and in-the-wild long footage, our framework outperforms prior diffusion-based methods in both fine-detail accuracy and temporal consistency, and uniquely maintains consistency across chunk boundaries on long videos.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Security token, Computer science, Inference, Generalization, Consistency (knowledge bases), Artificial intelligence

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange