OMMO dataset
收藏资源简介:
OMMO数据集是由复旦大学和腾讯PCG联合创建的大型户外多模态数据集,包含33个复杂场景,配有校准图像、点云和文本提示注释。数据集通过收集和精选大量真实航拍视频,设计质量审查模块,自动评估结合人工审查,移除低质量帧和校准失败的场景。志愿者还为每个场景和关键帧添加了文本描述。与现有数据集相比,OMMO数据集包含丰富的真实世界城市和自然场景,具有多种尺度、相机轨迹和光照条件,适用于新颖视图合成、表面重建和多模态NeRF等任务,旨在推动大规模户外场景的NeRF研究。
The OMMO Dataset is a large-scale outdoor multimodal dataset jointly developed by Fudan University and Tencent PCG. It encompasses 33 complex scenarios, with calibrated images, point clouds and text prompt annotations as accompanying data. The dataset is constructed via collecting and curating a large volume of real aerial videos, where a quality review module is designed, and low-quality frames and scenarios with failed calibration are removed through a combination of automatic evaluation and manual review. Volunteers have additionally added text descriptions for each scenario and key frame. Compared with existing datasets, the OMMO Dataset features rich real-world urban and natural scenarios with diverse scales, camera trajectories and lighting conditions. It is suitable for tasks such as novel view synthesis, surface reconstruction and multimodal NeRF, and aims to advance NeRF research on large-scale outdoor scenes.
- 1A Large-Scale Outdoor Multi-modal Dataset and Benchmark for Novel View Synthesis and Implicit Scene Reconstruction复旦大学 腾讯PCG · 2023年



