RegionReasoner-7B

Region-Grounded Multi-Round Visual Reasoning

ICLR 2026

Wenfang Sun · Hao Chen · Yingjun Du · Yefeng Zheng · Cees G. M. Snoek

---

RegionReasoner-7B is the released model for RegionReasoner, a reinforcement learning framework for region-grounded, multi-round visual reasoning. The framework combines explicit bounding-box references with global–local semantic consistency, and is evaluated on detection and segmentation tasks using RegionDial-Bench.

Abstract

Large vision-language models have achieved remarkable progress in visual reasoning, yet most existing systems rely on single-step or text-only reasoning, limiting their ability to iteratively refine understanding across multiple visual contexts. To address this limitation, we introduce a new multi-round visual reasoning benchmark with training and test sets spanning both detection and segmentation tasks, enabling systematic evaluation under iterative reasoning scenarios.

We further propose RegionReasoner, a reinforcement learning framework that enforces grounded reasoning by requiring each reasoning trace to explicitly cite the corresponding reference bounding boxes, while maintaining semantic coherence via a global-local consistency reward. This reward extracts key objects and nouns from both global scene captions and region-level captions, aligning them with the reasoning trace to ensure consistency across reasoning steps.

RegionReasoner is optimized with structured rewards combining grounding fidelity and global-local semantic alignment. Experiments on detection and segmentation tasks show that RegionReasoner-7B, together with our newly introduced benchmark RegionDial-Bench, considerably improves multi-round reasoning accuracy, spatial grounding precision, and global-local consistency, establishing a strong baseline for this emerging research direction.

Abstract reproduced from the paper by Wenfang Sun, Hao Chen, Yingjun Du, Yefeng Zheng, and Cees G. M. Snoek, under CC BY 4.0. Only paragraph formatting has been adjusted.

Resources

Resource	Link
Paper	arXiv:2602.03733
PDF	Read the paper
Code and evaluation instructions	lmsdss/RegionReasoner
Training dataset	lmsdss/regionreasoner_data
Test dataset	lmsdss/regionreasoner_test_data

Citation

@inproceedings{sun2026regionreasoner,
  title     = {RegionReasoner: Region-Grounded Multi-Round Visual Reasoning},
  author    = {Sun, Wenfang and Chen, Hao and Du, Yingjun and Zheng, Yefeng and Snoek, Cees G. M.},
  booktitle = {International Conference on Learning Representations},
  year      = {2026},
  url       = {https://arxiv.org/abs/2602.03733}
}