JDAI-CV/fast-reid

Health v2 · maintenance only

Adoption not part of the score

3981 stars · 879 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

FastReID is a PyTorch-based research platform implementing state-of-the-art re-identification algorithms for persons, vehicles, and faces. It serves as both a training toolbox with SoTA baselines and a library supporting projects like image retrieval, attribute recognition, and TensorRT deployment.

Use cases

• train a person re-identification model on Market-1501
• build an image retrieval system with re-ranking
• benchmark SoTA reid baselines in PyTorch
• convert reid models to ONNX or TensorRT for deployment
• recognize person attributes from images
• train face recognition models with partial FC
• run distributed multi-GPU training for reid

When to choose

• you need state-of-the-art re-identification baselines with reproducible configs
• you want a research platform extensible to retrieval, face recognition, and attribute tasks
• you need fast training with mixed precision, distributed GPUs, and TensorRT export

When to avoid

• you need a plug-and-play production reid API without model training
• you work outside PyTorch (e.g., TensorFlow or JAX)
• you need actively developed features - the project is in maintenance mode

Facets

library · maturity maintenance

machine-learning deep-learning image-processing computer-vision search-engine benchmarking computer-vision deep-learning machine-learning image-processing python re-identification person-reid pytorch image-retrieval face-recognition tensorrt onnx vehicle-reid toolbox gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="JDAI-CV/fast-reid")