SAM 3
The third-generation iteration of Meta AI's Segment Anything Model (SAM), a general-purpose foundation model for image and video segmentation. This version incorporates enhancements in video tracking and three-dimensional perception, enabling pixel-level segmentation masks to be generated for arbitrary objects in images or videos under near-zero-shot conditions.
Timeline (last 90 days)
Unsloth's Images page added a Detailer feature, which requires SAM 3 and a model that supports inpainting, used to redraw faces, hands or specified regions at higher resolution.
The pipeline uses SAM 3 for object segmentation and tracking, identifying fighters and separately extracting the ring surface and ropes as pixel-level masks.
All Facts (2)
Unsloth's Images page added a Detailer feature, which requires SAM 3 and a model that supports inpainting, used to redraw faces, hands or specified regions at higher resolution.
50%UnverifiedThe pipeline uses SAM 3 for object segmentation and tracking, identifying fighters and separately extracting the ring surface and ropes as pixel-level masks.
50%