facebookresearch/Detic

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2008 stars · 227 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Detic is the official code release for the ECCV 2022 paper 'Detecting Twenty-thousand Classes using Image-level Supervision'. It is an open-vocabulary object detector built on Detectron2 that uses CLIP embeddings to detect arbitrary classes given only class names, trained on ImageNet-21K with image-level labels.

Use cases

• detect arbitrary objects in images given class names
• run open-vocabulary object detection without finetuning
• train an object detector using only image-level labels
• evaluate zero-shot detection on LVIS, COCO, OpenImages, or Objects365
• detect objects from a custom vocabulary list
• run a detection demo in Colab or Hugging Face Spaces

When to choose

• you need to detect classes not covered by fixed-vocabulary detectors
• you want state-of-the-art open-vocabulary LVIS/COCO results
• you only have image-level labels rather than bounding box annotations
• you want a DETR-style detector with flexible vocabularies

When to avoid

• you need a lightweight detector for CPU-only or edge deployment
• you need a production system with active maintenance and support
• you only need standard fixed-vocabulary COCO detection
• you need a non-PyTorch or non-Detectron2 stack

Facets

library · maturity maintenance

computer-vision image-processing machine-learning deep-learning computer-vision machine-learning deep-learning artificial-intelligence python cross-platform object-detection open-vocabulary-detection clip detectron2 zero-shot-detection eccv-2022 research-code gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/Detic")