SAM-MT: Real-Time Multi-Target Video Segmentation

🛠 SAM-MT: Real-Time Multi-Target Video Segmentation

SAM-MT has been introduced—a framework based on the SAM2 architecture that allows for the simultaneous tracking of multiple objects using simple clicks. The model provides high processing speeds (36+ FPS with 10 targets on an RTX A6000) and efficient video memory usage.

🌍 SAM-MT solves the scalability problem of existing models, where latency increases alongside the number of objects.

👤 Segmentation tools are becoming faster and "smarter" when working with crowds. It is now possible to segment many objects simultaneously in real time.

Source 1: https://henghuiding.com/SAM-MT/ Source 2: https://github.com/FudanCVL/SAM-MT