Pipeline
From smartphone capture to neural scene representations
A High-Fidelity Dataset for 3D Reconstruction, Neural Rendering, and Spatial AI
OverMaps-Smartphone is a large-scale, high-fidelity collection of real-world spatial environments captured with consumer mobile devices. Each of the 171,635 scenes contains high-resolution multi-view imagery, metric-scale camera poses from ARKit/ARCore, COLMAP reconstructions, pre-computed 3D Gaussian Splatting models, depth maps, and rich semantic annotations — all processed with privacy-preserving pipelines.
86M+ high-resolution JPEG images with preserved EXIF metadata across 172K diverse real-world scenes.
Precise 4×4 camera-to-world transforms at metric scale via ARKit/ARCore SLAM.
Pre-computed 3DGS models for every scene, ready for immediate novel view synthesis.
Refined sparse and dense geometry with NetVLAD + ALIKED + SuperGlue + pixSfM pipeline.
Scene categories, weather, lighting, crowd density, and per-image captions via Qwen3-VL.
Automated person, vehicle, and license plate detection with YOLOv6 and inpainting masks.
From smartphone capture to neural scene representations
Dataset distribution and annotation insights
| # | Country | Scenes | Share |
|---|
Choose the dataset version that suits your research needs
172K scenes / 86M images. Complete dataset with all modalities: images, poses, 3DGS, COLMAP, depth, and annotations.
~672 TB Request Access1,000 curated scenes for quick experiments, benchmarking, and reproducing paper results.
~3,9 TB 🤗 DownloadA companion dataset of real-world scenes captured with 360° cameras for panoramic 3D reconstruction and immersive neural rendering.
TBA Stay TunedIf you use OverMaps-1K in your research, please cite our paper
@misc{OverMaps1k,
author = {OverTheReality},
title = {{OverMaps-1K Dataset}},
howpublished = {Hugging Face Datasets},
url = {https://huggingface.co/datasets/OverTheReality/OverMaps_1k},
year = {2025},
}