[image] 
[image] 
งาม飞思 Okay, SAM 3 (Segment Anything Model 3) by Meta is out (released late 2025 according to the context). It introduces "promptable concept segmentation" (open-vocabulary segmentation using text, like "yellow school bus"). It unifies image and video segmentation. It seems to use a "shared perception encoder (PE) vision backbone", and a "detector and tracker". The detector part....