[KongchangAI]
Model

MatteFormer

An image and video matting model based on the Transformer architecture, used to separate the subject from arbitrary backgrounds and generate an alpha channel.

Timeline (last 90 days)

Nov 5

Traditional green-screen matting relies on a solid-color background, whereas matting for arbitrary generated videos (such as depth-guided RVM, MatteFormer, etc.) needs to handle hair strands, semi-transparent edges, and motion blur.

Unverified50%

All Facts (1)

Source Articles