arXiv 2505.17166

ViDoRe Benchmark V2: Raising the Bar for Visual Retrieval

By Quentin Macé, António Loison, et al.

Published 2025-05-22

Citation lineage

Review the prior work and downstream research connected to this paper.

The ViDoRe Benchmark V1 was approaching saturation with top models exceeding 90% nDCG@5, limiting its ability to discern improvements. ViDoRe Benchmark V2 introduces realistic, challenging retrieval scenarios via blind contextual querying, long and cross-document queries, and a hybrid synthetic and human-in-the-loop query generation process. It comprises four diverse, multilingual datasets and provides clear evaluat…

View the original paper on arXiv