Paper Detail
Yangfu Li, Yuning Gong, Hongjian Zhan, Teng Li, Yuanhuiyi Lyu, Tianyi Chen, Qi Liu, Ziyuan Huang, Zhihang Zhong, Dandan Zheng, Yue Lu
Despite the success of Large-Vision Language Models (LVLMs), general optimization objectives (e.g., standard MLE) fail to constrain visual trajectories, leading to language bias and hallucination. To mitigate this, current methods introduce geometric priors from visual experts as additional supervision. However, we observe that such supervision is typically suboptimal: it is biased toward geometric precision and offers limited reasoning utility. To bridge this gap, we propose Perceptual Flow Network (PFlowNet), which eschews rigid alignment with the expert priors and achieves interpretable yet more effective visual reasoning. Specifically, PFlowNet decouples perception from reasoning to establish a self-conditioned generation process. Based on this, it integrates multi-dimensional rewards with vicinal geometric shaping via variational reinforcement learning, thereby facilitating reasoning-oriented perceptual behaviors while preserving visual reliability. PFlowNet delivers a provable performance guarantee and competitive empirical results, particularly setting new SOTA records on V* Bench (90.6%) and MME-RealWorld-lite (67.0%).
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{li2026perceptual,
title = {Perceptual Flow Network for Visually Grounded Reasoning},
author = {Yangfu Li and Yuning Gong and Hongjian Zhan and Teng Li and Yuanhuiyi Lyu and Tianyi Chen and Qi Liu and Ziyuan Huang and Zhihang Zhong and Dandan Zheng and Yue Lu},
year = {2026},
abstract = {Despite the success of Large-Vision Language Models (LVLMs), general optimization objectives (e.g., standard MLE) fail to constrain visual trajectories, leading to language bias and hallucination. To mitigate this, current methods introduce geometric priors from visual experts as additional supervision. However, we observe that such supervision is typically suboptimal: it is biased toward geometric precision and offers limited reasoning utility. To bridge this gap, we propose Perceptual Flow Net},
url = {https://huggingface.co/papers/2605.02730},
keywords = {Large-Vision Language Models, geometric priors, visual trajectories, language bias, hallucination, perceptual flow network, self-conditioned generation, variational reinforcement learning, multi-dimensional rewards, visual reasoning, V* Bench, MME-RealWorld-lite, huggingface daily},
eprint = {2605.02730},
archiveprefix = {arXiv},
}
{}