# cure-lab/MagicDrive

[ICLR24] Official implementation of the paper “MagicDrive: Street View Generation with Diverse 3D Geometry Control”

Repository: https://github.com/cure-lab/MagicDrive
Canonical: https://ross.abutalabs.com/products/magicdrive
Homepage: https://gaoruiyuan.com/magicdrive/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: autonomous-vehicles, deep-learning, image-generation, pytorch, diffusion-models, video-generation
Last push: 2025-04-21T10:20:04+00:00

## Health v2 (maintenance only)
Score: 35/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 17, release rhythm 35, longevity 75
- inputs: {"age_days": 1057, "days_push": 499, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_readme
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1166, forks 52 (observed 2026-08-28T04:03:50.315609+00:00)

## What it is
MagicDrive is the official PyTorch implementation of an ICLR 2024 paper for controllable street view generation using diffusion models with diverse 3D geometry control (BEV layouts, camera poses, text). It supports both image and video generation for autonomous driving scenarios and works with UNet and DiT architectures.

## Use cases
- generate street view images from bird's-eye view layouts
- synthesize multi-camera driving videos for 3d detection training
- create augmented autonomous driving datasets with edited scenes
- generate night or rainy variants of driving scenes
- research controllable diffusion models for 3d perception

## When to choose
- you need synthetic multi-camera street view data conditioned on 3D BEV geometry
- you are researching controllable diffusion generation for autonomous driving
- you want to augment nuScenes-style datasets with edited or rare scenarios

## When to avoid
- you need general-purpose image generation without 3D/BEV conditioning
- you lack GPU resources for training or running large diffusion models
- you need a production-ready data pipeline rather than a research codebase

## Facets
- artifact type: library
- maturity: active
- function: image-processing, video-processing, machine-learning, deep-learning, data-generation
- domain: autonomous-vehicles, deep-learning, computer-vision, artificial-intelligence, simulation
- platform: python
- tags: diffusion-models, street-view-generation, 3d-geometry-control, bev, data-synthesis, iclr24, pytorch, gpu, linux

## Member repositories
- cure-lab/MagicDrive (main) score 35

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:50.315609+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:28:58.983469+00:00, confidence not recorded.
  - homepage: https://gaoruiyuan.com/magicdrive/ (fetched 2026-08-29T12:35:12.502306+00:00, sha bdb23c5487be)
- Data as of 2026-08-30T08:39:29.467469+00:00.