ConsistentRFT: Reducing Visual Hallucinations in Flow-based Reinforcement Fine-Tuning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Tan, Xiaofeng, Liu, Jun, Fan, Yuanting, Gao, Bin-Bin, Jiang, Xi, Chen, Xiaochen, Peng, Jinlong, Wang, Chengjie, Wang, Hongsong, Zheng, Feng
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866915770573783040
author Tan, Xiaofeng
Liu, Jun
Fan, Yuanting
Gao, Bin-Bin
Jiang, Xi
Chen, Xiaochen
Peng, Jinlong
Wang, Chengjie
Wang, Hongsong
Zheng, Feng
author_facet Tan, Xiaofeng
Liu, Jun
Fan, Yuanting
Gao, Bin-Bin
Jiang, Xi
Chen, Xiaochen
Peng, Jinlong
Wang, Chengjie
Wang, Hongsong
Zheng, Feng
contents Reinforcement Fine-Tuning (RFT) on flow-based models is crucial for preference alignment. However, they often introduce visual hallucinations like over-optimized details and semantic misalignment. This work preliminarily explores why visual hallucinations arise and how to reduce them. We first investigate RFT methods from a unified perspective, and reveal the core problems stemming from two aspects, exploration and exploitation: (1) limited exploration during stochastic differential equation (SDE) rollouts, leading to an over-emphasis on local details at the expense of global semantics, and (2) trajectory imitation process inherent in policy gradient methods, distorting the model's foundational vector field and its cross-step consistency. Building on this, we propose ConsistentRFT, a general framework to mitigate these hallucinations. Specifically, we design a Dynamic Granularity Rollout (DGR) mechanism to balance exploration between global semantics and local details by dynamically scheduling different noise sources. We then introduce a Consistent Policy Gradient Optimization (CPGO) that preserves the model's consistency by aligning the current policy with a more stable prior. Extensive experiments demonstrate that ConsistentRFT significantly mitigates visual hallucinations, achieving average reductions of 49\% for low-level and 38\% for high-level perceptual hallucinations. Furthermore, ConsistentRFT outperforms other RFT methods on out-of-domain metrics, showing an improvement of 5.1\% (v.s. the baseline's decrease of -0.4\%) over FLUX1.dev. This is \href{https://xiaofeng-tan.github.io/projects/ConsistentRFT}{Project Page}.
format Preprint
id arxiv_https___arxiv_org_abs_2602_03425
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ConsistentRFT: Reducing Visual Hallucinations in Flow-based Reinforcement Fine-Tuning
Tan, Xiaofeng
Liu, Jun
Fan, Yuanting
Gao, Bin-Bin
Jiang, Xi
Chen, Xiaochen
Peng, Jinlong
Wang, Chengjie
Wang, Hongsong
Zheng, Feng
Computer Vision and Pattern Recognition
Reinforcement Fine-Tuning (RFT) on flow-based models is crucial for preference alignment. However, they often introduce visual hallucinations like over-optimized details and semantic misalignment. This work preliminarily explores why visual hallucinations arise and how to reduce them. We first investigate RFT methods from a unified perspective, and reveal the core problems stemming from two aspects, exploration and exploitation: (1) limited exploration during stochastic differential equation (SDE) rollouts, leading to an over-emphasis on local details at the expense of global semantics, and (2) trajectory imitation process inherent in policy gradient methods, distorting the model's foundational vector field and its cross-step consistency. Building on this, we propose ConsistentRFT, a general framework to mitigate these hallucinations. Specifically, we design a Dynamic Granularity Rollout (DGR) mechanism to balance exploration between global semantics and local details by dynamically scheduling different noise sources. We then introduce a Consistent Policy Gradient Optimization (CPGO) that preserves the model's consistency by aligning the current policy with a more stable prior. Extensive experiments demonstrate that ConsistentRFT significantly mitigates visual hallucinations, achieving average reductions of 49\% for low-level and 38\% for high-level perceptual hallucinations. Furthermore, ConsistentRFT outperforms other RFT methods on out-of-domain metrics, showing an improvement of 5.1\% (v.s. the baseline's decrease of -0.4\%) over FLUX1.dev. This is \href{https://xiaofeng-tan.github.io/projects/ConsistentRFT}{Project Page}.
title ConsistentRFT: Reducing Visual Hallucinations in Flow-based Reinforcement Fine-Tuning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2602.03425