A 360° Video Dataset for 3D Reconstruction, Neural Rendering, and Embodied AI
OverMaps-360Cam is a collection of real-world walking captures recorded with Insta360 X5 dual-fisheye 360° cameras at 8K. Its 2,712 scenes span 29 countries and 125 cities, totalling over 500 hours of capture and 2,200 km of walked paths. Each processed scene ships with dual-fisheye frames, perspective views with COLMAP poses, 3D Gaussian Splatting models, textured meshes, walkable navmeshes, and rich semantic annotations — all processed with privacy-preserving pipelines.
Native dual-fisheye capture with 3840×3840 frames per lens, plus the camera's IMU and a GPS track for every walk.
~12K rectified 1600×1600 perspective views per scene with sparse reconstructions and dense point clouds.
Pre-computed 5M-Gaussian 3DGS models (.ply and .splat) for immediate novel view synthesis.
Textured GLB/USDZ meshes, a collision mesh in three LOD tiers (1/3/10 cm error), metric floor alignment, and walkable navmeshes certified with Isaac Sim and MuJoCo physics tests.
Scene categories, weather, lighting, time of day, crowd density, and natural-language captions for 2.5K scenes.
Censored 360° recordings with per-frame privacy masks for every fisheye and perspective view.
Explore scenes from the dataset
From 360° camera capture to metric 3D deliverables
Choose the dataset version that suits your research needs
2.7K scenes / 500+ hours of 8K dual-fisheye 360° video. Complete dataset with all modalities: fisheye frames, perspective COLMAP, 3DGS, meshes, navmeshes, and annotations.
~60 TB Request Access50 curated scenes with every modality, for quick experiments and benchmarking.
~1.1 TB 🤗 DownloadIf you use OverMaps360-50 in your research, please cite our dataset
@misc{OverMaps360_50,
author = {OverTheReality},
title = {{OverMaps360-50 Dataset}},
howpublished = {Hugging Face Datasets},
url = {https://huggingface.co/datasets/OverTheReality/OverMaps360_50},
year = {2026},
}