[{"data":1,"prerenderedAt":12},["ShallowReactive",2],{"stack-tr-en-segment-anything":3},{"tagline":4,"description":5,"strengths":6,"useCase":11},"Meta's foundation model that segments any object from a single prompt.","Segment Anything is Meta's foundation model for promptable segmentation: a point, a box or a mask is enough to get a precise outline of an object never seen in training, with no fine-tuning. SAM 2 extends the approach to video with a memory that propagates and tracks masks from frame to frame, and SAM 3 accepts text prompts to segment every instance of a concept. It's our choice for speeding up dataset annotation or cutting out objects without training a dedicated model.",[7,8,9,10],"Zero-shot segmentation: point, box or mask in, precise outline out.","SAM 2: real-time mask tracking and propagation across video frames.","SAM 3: text prompts to segment every instance of a concept.","Open weights, PyTorch, ONNX export, chains with a detector such as YOLO.","Semi-automatic dataset annotation, e-commerce product cutouts, video object tracking, medical or agricultural imaging.",1792675979880]