megvii-research/PETR

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1065 stars · 160 forks observed · 2026-08-28

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

Official PyTorch implementation of PETR and PETRv2, transformer-based frameworks for multi-view 3D object detection and BEV segmentation from multi-camera images using 3D position embedding transformation. It serves as a strong research baseline for camera-only 3D perception, achieving state-of-the-art results on nuScenes.

Use cases

• multi-view 3d object detection from camera images
• bev segmentation from multi-camera setup
• research baseline for camera-only 3d perception
• temporal modeling for 3d detection
• 3d lane detection experiments
• reproduce nuscenes benchmark results

When to choose

• you need a proven camera-only 3D detection baseline for research
• you want state-of-the-art multi-camera detection or BEV segmentation on nuScenes
• you are building on position-embedding-based 3D perception methods

When to avoid

• you need a production-ready perception stack for a deployed vehicle
• you require a permissive license (license is non-standard)
• you need active feature development rather than a research artifact

Facets

library · maturity maintenance

machine-learning deep-learning computer-vision computer-vision autonomous-vehicles deep-learning artificial-intelligence python 3d-object-detection multi-camera position-embedding bev-segmentation nuscenes research-code transformer-detection linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="megvii-research/PETR")