ViDoRe
A visual document retrieval benchmark proposed by Illuin Tech, using a ColBERT-style late interaction mechanism to evaluate model retrieval over image-based documents such as scans, slides and posters. It provides multiple versions including v1, v2 and v3.
Timeline (last 90 days)
nanoMTEB is a streamlined subset of MTEB used for low-cost evaluation of the text-matching capability of retrieval and reranking models; ViDoRe is a benchmark for visual document retrieval, and V3 is its newer version.
ViDoRe is a visual document retrieval benchmark proposed by Illuin Tech, short for Visual Document Retrieval, specifically evaluating models' retrieval capability on image-based documents, and provides versions such as v1, v2, and v3.
ViDoRe adopts a ColBERT-style late interaction mechanism: documents are split into patches, each patch generates a dense vector, queries are likewise encoded as multi-vectors, and scoring is done via MaxSim by summing the per-vector maximum similarities.
All Facts (3)
nanoMTEB is a streamlined subset of MTEB used for low-cost evaluation of the text-matching capability of retrieval and reranking models; ViDoRe is a benchmark for visual document retrieval, and V3 is its newer version.
50%UnverifiedViDoRe is a visual document retrieval benchmark proposed by Illuin Tech, short for Visual Document Retrieval, specifically evaluating models' retrieval capability on image-based documents, and provides versions such as v1, v2, and v3.
50%UnverifiedViDoRe adopts a ColBERT-style late interaction mechanism: documents are split into patches, each patch generates a dense vector, queries are likewise encoded as multi-vectors, and scoring is done via MaxSim by summing the per-vector maximum similarities.
50%