MegaDepth-X: Large-Scale Internet Photo Reconstruction Dataset

MegaDepth-X is a large-scale dataset for long-tail 3D reconstruction from Internet photo collections. The dataset, created by Yuan Li and others, addresses the challenge of highly long-tailed photo distributions where a small number of landmarks have many photos while most have few. It was published in 2024 and is hosted on Hugging Face.

Use Cases

• Training 3D reconstruction models based on large-scale Internet photo collections.
• Developing depth estimation algorithms for long-tail visual data.
• Benchmarking photogrammetry methods on real-world, unstructured imagery.
• Researching methods to handle data sparsity in computer vision tasks.

Strengths

• Dataset is described as 'large-scale' for 3D reconstruction.
• Specifically designed to address the 'long-tail' distribution problem in Internet photos.
• Associated with a peer-reviewed paper published on arXiv in 2024.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.
• Data may reflect geographic or source bias inherent to its collection from Internet photos.

Provenance

Source

Internet photo collections, as per the description.

Collection Method

Photo reconstruction, likely via structure-from-motion or similar techniques.

Time Range

null

Freshness

Last updated 2026-05-07 19:30:53; freshness should be verified.

Geography

null