Kuaishou Audio & Video Technology
Jun 17, 2021 · Artificial Intelligence
How Deep Learning Revolutionizes Video Matting: A Two‑Stage Framework
Leveraging deep neural networks, this article introduces a pioneering two‑stage video matting framework that propagates sparse Trimap annotations across frames using cross‑attention, integrates spatio‑temporal features via a ST‑FAM module, and demonstrates superior performance on synthetic and real HD video datasets.
Computer Visiondeep learningspatio-temporal fusion
0 likes · 10 min read