AgentMatting: Boosting context aggregation for image matting with context agent

Jide Li, Kequan Yang, Chao Yin, Xiaoqiang Li

Abstract: Highlights•We propose AgentMatting, a matting model that balances context and local details.•We introduce the context agent and the CAG block to enhance context aggregation.•Experiments show AgentMatting excels across multiple datasets.•AgentMatting is more computationally efficient than ViT-based methods.