aiqbench analysis
R3G introduces a modular framework that enhances visual question answering by explicitly planning reasoning steps before retrieving and reranking relevant images. This approach demonstrates state-of-the-art performance across multiple multimodal large language model backbones, effectively bridging the gap between visual evidence selection and logical inference. By decoupling retrieval from reasoning, the method offers a scalable architecture for improving accuracy in vision-centric tasks without relying on single-model architectures.
R3G: A Reasoning-Retrieval-Reranking Framework for Vision-Centric Answer Generation
arXiv:2602.00104v4 Announce Type: replace-cross Abstract: Vision-centric retrieval for VQA requires retrieving images to supply missing visual cues and integrating them into the reasoning process. However, selecting the right images and integrating them effectively into the model's reasoning remains challenging. To address this challenge, we propose R3G, a modular Reasoning-Retrieval-Reranking framework. It first produces a brief reasoning plan that specifies the required visual cues, then adopts a two-stage strategy, with coarse retrieval followed by fine-grained reranking, to select evidence images. On MRAG-Bench, R3G improves accuracy across six MLLM backbones and nine sub-scenarios, achieving state-of-the-art overall performance. Ablations show that sufficiency-aware reranking and reasoning steps are complementary, helping the model both choose the right images and use them well. We release code and data at https://github.com/czh24/R3G.
Abstract:Vision-centric retrieval for VQA requires retrieving images to supply missing visual cues and integrating them into the reasoning process. However, selecting the right images and integrating them effectively into the model's reasoning remains challenging. To address this challenge, we propose R3G, a modular Reasoning-Retrieval-Reranking framework. It first produces a brief reasoning plan that specifies the required visual cues, then adopts a two-stage strategy, with coarse retrieval followed by fine-grained reranking, to select evidence images. On MRAG-Bench, R3G improves accuracy across six MLLM backbones and nine sub-scenarios, achieving state-of-the-art overall performance. Ablations show that sufficiency-aware reranking and reasoning steps are complementary, helping the model both choose the right images and use them well. We release code and data at this https URL.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2602.00104 [cs.CV]
(or arXiv:2602.00104v4 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2602.00104 Focus to learn more arXiv-issued DOI via DataCite
ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 8602-8606, 2026
https://doi.org/10.1109/ICASSP55912.2026.11463453 Focus to learn more DOI(s) linking to related resources
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 246 words
Related coverage
R3G: A Reasoning-Retrieval-Reranking Framework for Vision-Centric Answer Generation
arXiv · Jun 4, 2026
From Terminology to Diagrams: Visual-Instruction Generation for Scientific Diagram Understanding
arXiv · Sep 2, 2026
NVIDIA GTC Taipei at COMPUTEX: Live Updates on What’s Next in AI
NVIDIA · May 21, 2026
PixSearch: Region-Grounded Retrieval for Knowledgeable Large Multimodal Models
arXiv · Oct 15, 2026