MAD-Cars
收藏资源简介:
MAD-Cars数据集是由Yandex Research、HSE University和Skoltech共同创建的,包含约70,000个360度车辆视频,涵盖了多种品牌、型号、颜色和光照条件,用于自动驾驶场景的重建和模拟。该数据集为自动驾驶领域的研究提供了重要的数据支持,有助于提高自动驾驶系统的感知和规划能力。
The MAD-Cars dataset was co-created by Yandex Research, HSE University and Skoltech. It comprises approximately 70,000 360-degree vehicle videos covering various vehicle brands, models, colors and lighting conditions, and is designed for autonomous driving scenario reconstruction and simulation. This dataset offers crucial data support for research in the autonomous driving domain, and contributes to enhancing the perception and planning capabilities of autonomous driving systems.
MADrive: Memory-Augmented Driving Scene Modeling
概述
- 数据集名称: MAD-Cars
- 开发团队: Yandex Research, HSE University, Skoltech
- 核心贡献: 提出MADrive框架,通过外部记忆库增强现有场景重建方法,支持显著改变或新颖驾驶场景的光照真实合成
- 数据集规模: 约70,000个360°汽车视频
- 技术特点: 支持车辆多视角完整表示、方向对齐和重光照
方法
- 检索模块: 从外部数据库查找相似车辆
- 重建流程:
- 从检索视频生成详细3D车辆模型
- 使用可重光照的2D高斯泼溅表示车辆
- 应用不透明度掩模去除背景泼溅
- 通过外部法线贴图正则化模型几何
- 场景集成: 调整车辆光照条件并与背景合成
评估场景
- 轨迹外推: 生成未来车辆外观
- 新轨迹生成: 支持原始轨迹和修改轨迹的视频生成
- 重光照对比: 展示带/不带重光照效果的对比案例
学术引用
bibtex @article{karpikova2025madrivememoryaugmenteddrivingscene, title={MADrive: Memory-Augmented Driving Scene Modeling}, author={Polina Karpikova and Daniil Selikhanovych and Kirill Struminsky and Ruslan Musaev and Maria Golitsyna and Dmitry Baranchuk}, year={2025}, eprint={2506.21520}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2506.21520} }
资源链接
- arXiv论文: https://arxiv.org/abs/2506.21520

- 1MADrive: Memory-Augmented Driving Scene ModelingYandex Research, HSE University, Skoltech · 2025年



