OACIRR Benchmark for Referential Anchoring in Image Retrieval

OACIRR is a benchmark dataset for referential anchoring in composed image retrieval, released by HaHaJun1101. It supports research on tasks beyond semantic search, as presented in a CVPR 2026 paper. The dataset was officially released on March 24, 2026.

Use Cases

• Train models for referential anchoring tasks using the benchmark's defined image and query structures.
• Evaluate retrieval performance on composed queries involving specific referential features.
• Benchmark model checkpoints like AdaFocal against the dataset's defined evaluation protocols.

Strengths

• Dataset is associated with a peer-reviewed CVPR 2026 paper.
• Officially released on March 24, 2026, indicating recent availability.

Limitations

• Key dataset specifications like row count, column names, and file formats are unknown.
• Sample data is unavailable, making initial inspection and understanding difficult.

Provenance

Source

HaHaJun1101 on Hugging Face.

Collection Method

Created as a benchmark for the research paper 'Beyond Semantic Search: Towards Referential Anchoring in Composed Image Retrieval'.

Freshness

Last updated on 2026-03-24.