SOCO: Semantic Object Correspondence Dataset with Keypoint Annotations

SOCO is a semantic object correspondence dataset with images, per-view keypoint annotations, and pair annotations for evaluating visual features. The dataset was created by GenIntelLab and was last updated on the platform in June 2026. Evaluation code is available in the associated OmniProbe repository.

Use Cases

• Evaluating visual feature representations based on semantic object correspondence annotations.
• Training models for keypoint detection and matching based on per-view keypoint annotations.
• Benchmarking image matching algorithms based on provided intra and cross-category pair annotations.

Strengths

• Includes pair annotations for evaluation across intra and cross-category splits.
• Provides per-view keypoint annotations for detailed object correspondence.

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

GenIntelLab

Freshness

Last updated 2026-06-01 18:03:20; freshness should be verified.