ECom-RF-IMMR: Four Image-to-Multimodal Item Retrieval Benchmarks

Four image-to-multimodal item retrieval datasets are contained in this repository. The collection includes two newly constructed evaluation datasets and two adapted from public e-commerce benchmarks. The datasets were created by author xyxy01 and last updated on 2026-05-27.

Use Cases

• Benchmarking image-to-multimodal retrieval models based on the described evaluation datasets.
• Training models for e-commerce product search based on image region queries.
• Adapting public e-commerce retrieval benchmarks for new research tasks.
• Evaluating retrieval performance on specialized datasets like those with mosaic image regions.

Strengths

• Contains four distinct datasets, providing multiple benchmarks for comparison.
• Includes two newly constructed datasets (ECom_RF_IMMR_Normal and ECom_RF_IMMR_Mosaic) specifically for the image-to-multimodal retrieval task.
• Last updated on 2026-05-27, suggesting recent maintenance.

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, file formats, and license information are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Two datasets are constructed by the author, and two are adapted from public e-commerce benchmarks.

Freshness

Last updated 2026-05-27 09:35:12; freshness should be verified.