WideDepth: Millimeter-Accurate Benchmark for Fisheye Depth Estimation

Ilia Indyk¹, Ignat Penshin¹, Ivan Sosin¹, Maxim Monastyrny¹, Aleksei Valenkov¹, Ilya Makarov^{2, 3}

¹Robotics Center, ²AXXX, ³ISP RAS,
ICRA 2026

Paper Benchmark Train set Code

101 indoor scenes
5K fisheye stereo pairs
±1 mm depth precision at 10 m
120–195° field of view
18K LiDAR-labeled training pairs
up to −62% stereo error after fine‑tuning

WideDepth pipeline overview — Overview of our proposed pipeline for generating fisheye stereo RGB, depth, and disparity from LiDAR scans.

Abstract

Fisheye cameras are increasingly adopted in robotics for near-field manipulation, navigation, and immersive perception, yet indoor depth benchmarks with accurate ground truth are still missing. To address this, we introduce WideDepth — the first indoor dataset for fisheye depth estimation, featuring 101 scenes containing 5K high-resolution stereo pairs labeled with millimeter-level ground truth depth and disparity. Our dataset also includes paired pinhole and fisheye samples across varying fields of view and baselines in both horizontal and vertical stereo setups. We further propose a method to adapt pinhole-trained stereo models to fisheye images and introduce a novel stereo fisheye image generation pipeline based on high-resolution LiDAR scans. Leveraging these methods, we thoroughly evaluate state-of-the-art monocular depth, stereo matching, and depth completion models on our benchmark. Additionally, we provide 18K LiDAR-derived sparse depth training samples, achieving up to a 62% performance boost on fisheye data when fine-tuning pinhole-based stereo models. In summary, the high precision and versatility of our benchmark set a strong foundation for advancing research in fisheye depth estimation and robotics perception.

Benchmark results

Dataset comparison

Parameter	OmniScape	SynWoodScape	Oxford RobotCar	KITTI-360	WoodScape	WideDepth train (ours)	WideDepth benchmark (ours)
Real / synthetic	Synthetic	Synthetic	Real	Real	Real	Real	Real
Domain	Outdoor	Outdoor	Outdoor	Outdoor	Outdoor	Outdoor	Indoor
Fisheye resolution	1024×1024	1280×966	1024×1024	1400×1400	1280×966	1920×1080	2048×1152
Fisheye HFOV	185°	190°	180°	180°	190°	180°	120–195°
Precision at 10 m	Not reported	Not reported	±30 mm	±20 mm	±20 mm	±20 mm	±1 mm
Horizontal stereo	Yes	No	No	No	No	No	Yes
Vertical stereo	No	No	No	No	No	Yes	Yes
Pinhole	No	No	Yes	Yes	No	No	Yes
Dense depth map	Yes	Yes	No	No	No	No	Yes

Our benchmark offers unparalleled precision of depth maps, leveraging varying FOV and high-resolution images, surpassing all existing fisheye datasets. Our training dataset is the first-ever with a vertical stereo setup.

Stereo depth estimation

Stereo qualitative results — Qualitative results for stereo with FOV 195 using the StereoBase model. The model shows no degradation from geometric distortion, demonstrating the success of our approach in adapting pinhole models to fisheye data.

Monocular depth estimation

Monocular qualitative results — Comparison of the best and worst models on a 195-degree FOV. Despite its ability to distinguish small details well, Depth Anything V2 performs significantly worse in metric depth estimation on extremely wide angles compared to Depth Pro.

Video Presentation

Citation

If you use WideDepth in your research, please cite our ICRA 2026 paper.

@inproceedings{indyk2026widedepth,
  title={WideDepth: Millimeter-Accurate Benchmark for Fisheye Depth Estimation},
  author={Ilia Indyk and Ignat Penshin and Ivan Sosin and Maxim Monastyrny
          and Aleksei Valenkov and Ilya Makarov},
  booktitle={2026 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={17817--17824},
  year={2026},
  doi={10.1109/ICRA57385.2026.11696790}
}