Meta's research platform for detection and segmentation.
Detectron2 is the library a lot of published vision work is implemented against — Mask R-CNN, RetinaNet, panoptic segmentation and DensePose all live here with reference configs and a model zoo of trained weights. It assumes familiarity with the literature, and rewards it with a modular design where a new backbone or head slots in cleanly.
Real-time object detection you can fine-tune in an afternoon.
Visual builder for agents and RAG flows, deployable as an API.
Fair-code workflow automation with AI steps built in.
Track the experiments, register the models, ship the good one.
Fine-tuning driven entirely by a YAML config.
Fine-tune a hundred different models from one interface.
No reviews yet — be the first to review this project.
Sign in to write a review.
Sign In