Search papers, labs, and topics across Lattice.
Fudan University
2
0
2
SAM-MT achieves real-time multi-target video segmentation at over 36 FPS, decoupling latency from target count while preserving high-quality segmentation.
Achieving state-of-the-art video matting performance with a framework trained exclusively on images reveals the untapped potential of existing trackers in fine-grained tasks.
