Best Instance Segmentation Models

RF-DETR Seg, Mask R-CNN lineage models, and YOLO segmentation variants learn to segment your specific classes and run fast enough for production video and high-volume pipelines. Like detectors, they need labeled masks to train, but they then produce consistent instance masks with no prompting, which is what automated measurement, counting, and robotics pipelines need.

RF-DETR Seg is Apache 2.0 at every size, so it's commercially usable out of the box. Ultralytics YOLO segmentation models are AGPL-3.0, so commercial use needs a commercial license; deploying on Roboflow includes one for every user.

SAM-family models segment almost anything from a click, box, or, in newer variants, a text phrase, with no training. They are the fastest way to get pixel-accurate masks for labeling datasets, cutting objects out in editing tools, and one-off analysis. The tradeoff: they do not know your classes, so something else (a person, a detector, or a text prompt) has to decide what to segment.

The standard workflow uses a promptable segmenter to build a mask dataset quickly, then trains a compact segmentation model on those masks for production. You get SAM-quality boundaries in the dataset and millisecond inference in deployment.