Seoul World Model (SWM) Dataset
收藏资源简介:
首尔世界模型(SWM)数据集是由韩国科学技术院、NAVER AI实验室等机构联合构建的城市级仿真数据集,旨在实现真实城市环境下的视频生成。该数据集包含44万条首尔街景图像,辅以合成驾驶视频(12.7K条)及公开驾驶数据集,覆盖431,500平方米城市区域。数据通过跨时间配对策略消除瞬时内容干扰,并采用间歇冻结帧技术解决街景稀疏性问题。其核心应用为自动驾驶场景生成、城市规划和沉浸式导航,通过地理检索增强生成技术将视频内容锚定真实城市几何结构,支持数百米轨迹的连贯生成。
The Seoul World Model (SWM) Dataset is an urban-scale simulation dataset jointly developed by entities including the Korea Advanced Institute of Science and Technology (KAIST) and NAVER AI Lab, aimed at generating videos within realistic urban environments. This dataset comprises 440,000 street view images of Seoul, supplemented by 12.7K synthetic driving videos and public driving datasets, covering an urban area of 431,500 square meters. It eliminates transient content disturbances through a cross-temporal pairing strategy, and addresses the sparsity issue of street view data using the intermittent freeze-frame technique. Its core applications cover autonomous driving scenario generation, urban planning and immersive navigation. By leveraging geographic retrieval-augmented generation technology, it anchors video content to real-world urban geometric structures, enabling coherent generation of trajectories spanning hundreds of meters.




