Perceptual Flow Network for Visually Grounded Reasoning
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917458932137984 |
|---|---|
| author | Li, Yangfu Gong, Yuning Zhan, Hongjian Li, Teng Lyu, Yuanhuiyi Chen, Tianyi Liu, Qi Huang, Ziyuan Zhong, Zhihang Zheng, Dandan Lu, Yue |
| author_facet | Li, Yangfu Gong, Yuning Zhan, Hongjian Li, Teng Lyu, Yuanhuiyi Chen, Tianyi Liu, Qi Huang, Ziyuan Zhong, Zhihang Zheng, Dandan Lu, Yue |
| contents | 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%). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_02730 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Perceptual Flow Network for Visually Grounded Reasoning Li, Yangfu Gong, Yuning Zhan, Hongjian Li, Teng Lyu, Yuanhuiyi Chen, Tianyi Liu, Qi Huang, Ziyuan Zhong, Zhihang Zheng, Dandan Lu, Yue Computer Vision and Pattern Recognition Artificial Intelligence I.2.6; I.2.10; I.4.8 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%). |
| title | Perceptual Flow Network for Visually Grounded Reasoning |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence I.2.6; I.2.10; I.4.8 |
| url | https://arxiv.org/abs/2605.02730 |