UnrealMVS

A large-scale synthetic omnidirectional depth dataset rendered in Unreal Engine. Designed for training and evaluating multi-view stereo networks (e.g. OmniMVS) on fisheye camera rigs.

Each scene is captured with a 4-fisheye-camera rig mounted in a fixed configuration. Per frame the dataset provides:

File	Description
FisheyeCamera_1.exr – FisheyeCamera_4.exr	Raw fisheye images, float32
Equirectangular_Depth.exr	Ground-truth depth in metres, float32 (0 = invalid)
Equirectangular.exr	Equirectangular colour panorama, float32
calibration_intrinsic.json	Per-camera intrinsics (one file per scene)
calibration_extrinsic.json	Rig extrinsics — camera-to-rig transforms

The image below shows how the four fisheye cameras cover the full 360° sphere for one example frame (FactoryEnvironment): each coloured region is the contribution of one camera after projection onto the equirectangular panorama, with the equirectangular ground-truth depth blended in.

---

Scene overviews

Each image below shows a composited panoramic overview of one scene, generated from the fisheye images, intrinsics, extrinsics, and equirectangular depth map.

Bazaar 500 frames	Cambodia 500 frames
Catacomb 850 frames	Cigar Room 500 frames
CitySampleDay 3000 frames	CitySampleNight 2000 frames
ElectricDreamsEnv 900 frames	Engine Hall 500 frames
FactoryEnvironment 2000 frames	Mediterranean 500 frames
PolarStation 250 frames	PoliceHeadquarters 1000 frames
Tokyo 500 frames	WindyHouse 500 frames

---

Code & Tools (coming soon)

A companion GitHub repository for UnrealMVS will be published at a later date. It will include:

• Training pipeline — scripts for loading and preprocessing this dataset for training neural networks for depth estimation from multi-fisheye-camera rigs.
• Overview image generation — code showing how the overview images below were created: given per-camera intrinsics, extrinsics, raw fisheye images, and the equirectangular ground-truth depth, a full 360° panorama is composited.
• Dataset conversion script — a ready-to-run script that downloads a further public fisheye dataset and converts it into the same format used here, so it can be used as a drop-in replacement or supplement for training.

---

Dataset structure

FactoryEnvironment/
├── calibration_intrinsic.json
├── calibration_extrinsic.json
├── mask.png
├── 0000/
│   ├── FisheyeCamera_1.exr
│   ├── FisheyeCamera_2.exr
│   ├── FisheyeCamera_3.exr
│   ├── FisheyeCamera_4.exr
│   ├── Equirectangular_Depth.exr
│   └── Equirectangular.exr
├── 0001/
│   └── ...
└── ...

---

Acknowledgements

This dataset was created with the generous support of the Westfälische Hochschule Gelsenkirchen (Westphalian University of Applied Sciences) and the Technische Universität Dortmund. We thank both institutions for providing the resources and infrastructure that made this publication possible.

---