Multi-View Fisheye And Omnidirectional Images For 3D Reconstruction

A collection of scenes captured with multiple fisheye and 360° omnidirectional cameras for 3D reconstruction tasks. The dataset, created by youlenda, was last updated in April 2026. Each scene contains fisheye frames, corresponding masks, omnidirectional frames, and associated sparse COLMAP reconstructions.

Use Cases

• Train multi-view stereo depth estimation models using synchronized fisheye frames from camera1 and camera2.
• Develop 3D scene reconstruction pipelines by leveraging the provided sparse COLMAP point clouds in sparse/0/.
• Benchmark novel view synthesis methods using the paired omnidirectional frames in omni/images/.
• Evaluate foreground-background segmentation models on the fisheye masks provided in the masks/ directories.

Strengths

• Provides synchronized multi-view data with at least two distinct camera perspectives per scene.
• Includes both raw fisheye images and pre-computed sparse 3D reconstructions for each scene.
• Contains corresponding segmentation masks for fisheye frames, aiding in foreground isolation.

Limitations

• Unknown total number of scenes, rows, or images, making it difficult to assess scale.
• Unknown geographic and temporal coverage of the captured scenes.
• Data completeness and consistency across all scenes is unverified.

Provenance

Source

youlenda on Hugging Face.

Collection Method

Casual 360° captures, processed with COLMAP for sparse reconstruction.

Freshness

Last updated in April 2026.