Vidore V-2

The 'ViDoRe v2' dataset/benchmark contains a limited corpus used to evaluate the performance of multi-vector visual retrieval systems, specifically focusing on metrics like NDCG and Recall.

Papers using Vidore V-2 (2)

Visual RAG Toolkit: Scaling Multi-Vector Visual Retrieval with Training-Free Pooling and Multi-Stage Search2026ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System2026