metadata
pipeline_tag: image-text-to-text
V-Retrver: Evidence-Driven Agentic Reasoning for Universal Multimodal Retrieval
V-Retrver is an evidence-driven retrieval framework that reformulates multimodal retrieval as an agentic reasoning process grounded in visual inspection.
About V-Retrver
V-Retrver enables a Multimodal Large Language Model (MLLM) to selectively acquire visual evidence during reasoning via external visual tools. It performs a multimodal interleaved reasoning process that alternates between hypothesis generation and targeted visual verification.
To train this agent, the authors adopted a curriculum-based learning strategy combining:
- Supervised Reasoning Activation: Initial activation of retrieval-specific reasoning.
- Rejection-Based Refinement: Improving reasoning reliability via rejection sampling.
- Reinforcement Learning: Fine-tuning with an evidence-aligned objective.
Experiments across multiple benchmarks demonstrate significant improvements in retrieval accuracy (averaging 23.0%), perception-driven reasoning reliability, and generalization.
Resources
- Paper: V-Retrver: Evidence-Driven Agentic Reasoning for Universal Multimodal Retrieval
- GitHub Repository: https://github.com/chendy25/V-Retrver
Citation
If you find this work helpful, please consider citing:
@article{chen2026vretrver,
title={V-Retrver: Evidence-Driven Agentic Reasoning for Universal Multimodal Retrieval},
author={Chen, Dongyang and Wang, Chaoyang and Su, Dezhao and Xiao, Xi and Zhang, Zeyu and Xiong, Jing and Li, Qing and Shang, Yuzhang and Ka, Shichao},
journal={arXiv preprint arXiv:2602.06034},
year={2026}
}