V-Retrver: Evidence-Driven Agentic Reasoning for Universal Multimodal Retrieval

TL;DR

V-Retrver introduces an evidence-driven agentic reasoning framework for multimodal retrieval, enabling models to actively verify visual evidence and significantly improve retrieval accuracy and reasoning reliability.

Contribution

It presents a novel framework that allows multimodal models to actively gather visual evidence during reasoning, enhancing retrieval performance and interpretability.

Findings

23.0% average improvement in retrieval accuracy

Enhanced reasoning reliability and generalization

Effective evidence-gathering through curriculum-based training

Abstract

Multimodal Large Language Models (MLLMs) have recently been applied to universal multimodal retrieval, where Chain-of-Thought (CoT) reasoning improves candidate reranking. However, existing approaches remain largely language-driven, relying on static visual encodings and lacking the ability to actively verify fine-grained visual evidence, which often leads to speculative reasoning in visually ambiguous cases. We propose V-Retrver, an evidence-driven retrieval framework that reformulates multimodal retrieval as an agentic reasoning process grounded in visual inspection. V-Retrver enables an MLLM to selectively acquire visual evidence during reasoning via external visual tools, performing a multimodal interleaved reasoning process that alternates between hypothesis generation and targeted visual verification.To train such an evidence-gathering retrieval agent, we adopt a curriculum-based learning strategy combining supervised reasoning activation, rejection-based refinement, and reinforcement learning with an evidence-aligned objective. Experiments across multiple multimodal retrieval benchmarks demonstrate consistent improvements in retrieval accuracy (with 23.0% improvements on average), perception-driven reasoning reliability, and generalization.