Widedepth: Indoor Fisheye Depth-Estimation Benchmark with LiDAR Ground Truth

An indoor fisheye depth-estimation benchmark dataset with millimeter-accurate ground-truth depth and disparity rendered from high-resolution LiDAR scans. It includes 101 indoor scenes captured with a fixed camera configuration of 195° FOV and 300 mm focal length. The dataset was created by Voxel51 and released on HuggingFace in May 2026.

Use Cases

• Train depth-estimation models based on millimeter-accurate LiDAR ground truth.
• Benchmark stereo vision algorithms based on the provided disparity maps.
• Develop and test computer vision models for indoor robotic navigation based on fisheye imagery.
• Evaluate model performance on wide-field-of-view (195° FOV) image data.

Strengths

• Ground-truth depth and disparity are millimeter-accurate, derived from high-resolution LiDAR scans.
• Includes 101 distinct indoor scenes, providing a variety of environments.
• Uses a consistent, fixed camera configuration (195° FOV, 300 mm focal length) for all scenes.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Voxel51 via HuggingFace.

Collection Method

Ground-truth depth and disparity rendered from high-resolution LiDAR scans of indoor scenes.

Freshness

Last updated 2026-05-26 21:17:20; freshness should be verified.