3D-Mirage: 468 Scenes of Painting and Street-Art Illusions for Depth Model Testing

468 real-world scenes of painting and street-art illusions form a benchmark for probing 3D hallucinations in monocular depth models. The dataset backs the ECCV 2026 paper 'The 3D Mirage: Probing and Taming 3D Hallucinations' and is authored by 3dmirage. It was last updated on July 4, 2026.

Use Cases

• Benchmarking depth model robustness based on real-world optical illusions
• Probing 3D hallucination tendencies in models based on texture and prior trust
• Developing methods to 'tame' 3D hallucinations as referenced in the associated paper

Strengths

• Contains 468 real-world scenes, providing a concrete scale for evaluation
• Focuses on a specific, documented failure mode (3D mirages) for depth models
• Associated with a peer-reviewed ECCV 2026 paper, suggesting academic rigor

Limitations

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

Provenance

Source

huggingface

Collection Method

Likely collected from real-world scenes of painting and street-art illusions.

Freshness

Last updated 2026-07-04 03:25:13; freshness should be verified