Tator: local FastAPI CLIP+SAM annotator for faster YOLO labeling
Overview
Tator is a single-machine web app with a FastAPI backend that speeds up image annotation using CLIP for class suggestions and SAM for auto box refinement, with optional in-browser CLIP training and model management. It focuses on fast bounding-box workflows for YOLO-style datasets and keeps data fully local. The repo is under active development; dataset management and SAM-related training are WIP, so expect rough edges.
All Sources
Story Timeline
UPDATE Update: Tator
New: the UI now bundles labeling, CLIP training, and model management in-browser, plus fresh labeling modes like Auto Class Corrector, one-click point-to-box, and multi-point prompts. Tator also introduces early SAM3 support (sam3_local/sam3_lite) with recipe mining and training marked WIP, while dataset management remains rough. This moves beyond simple suggestions/refinement toward more automated, point-driven box creation and stricter auto-class correction.