αMatte4K & µMatting: Dataset and Model for Ultra-Micro Precision Alpha Video Matting

High-resolution human video matting aims to predict accurate alpha mattes for semi-transparent regions while ensuring temporal consistency across frames. Despite notable progress, current methods still fail to achieve a satisfactory trade-off between quality and efficiency, with limitations in subject stability, temporal modeling, and computational cost. In this paper, we introduce µMatting, an innovative resolution-agnostic two-stage framework for video matting: (1) coarse matte localization using a portrait-aware masked autoencoder;

(2) refinement of critical regions via sparse 3D convolution, augmented by a temporal modulator that injects global spatio-temporal cues for enhanced consistency and contextual awareness. From data perspective, existing research remains limited by the insufficient quality of datasets, including (1) inaccurate alpha fractional values resulting from imperfect annotation, and (2) visual inconsistencies arising from arbitrary foregroundbackground compositions that lack natural coherence. To address this, we introduce αMatte4K, a large-scale 4Kresolution human video matting dataset, which achieves accurate annotations and physical consistency through physically based rendering (PBR). Extensive experiments show that µMatting surpasses state-of-the-art methods in accuracy and spatio-temporal consistency, while αMatte4K boosts baseline performance, driving applications in realworld scenarios. The project is open-sourced at https: //github.com/kadatec/mu-Matting.