Anomaly-Preference Image Generation

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Fuyun, Wang, Yuanzhi, Guo, Xu, Huang, Sujia, Zhang, Tong, Wang, Dan, Yan, Hui, Liu, Xin, Cui, Zhen
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910231314825216
author Wang, Fuyun
Wang, Yuanzhi
Guo, Xu
Huang, Sujia
Zhang, Tong
Wang, Dan
Yan, Hui
Liu, Xin
Cui, Zhen
author_facet Wang, Fuyun
Wang, Yuanzhi
Guo, Xu
Huang, Sujia
Zhang, Tong
Wang, Dan
Yan, Hui
Liu, Xin
Cui, Zhen
contents Synthesizing realistic and diverse anomalous samples from limited data is vital for robust model generalization. However, existing methods struggle to reconcile fidelity and diversity, often hampered by distribution misalignment and overfitting, respectively.To mitigate this, we introduce Anomaly Preference Optimization,a novel paradigm that reformulates anomaly generation as a preference learning problem.Central to our approach is an implicit preference alignment mechanism that leverages real anomalies as positive references, deriving optimization signals directly from denoising trajectory deviations without requiring costly human annotation. Furthermore, we propose a Time-Aware Capacity Allocation module that dynamically distributes model capacity along the diffusion timeline,prioritizing structural diversity during highnoise phases while enhancing fine-grained fidelity in low-noise stages. During inference, a hierarchical sampling strategy modulates the coherencealignment trade-off, enabling precise control over generation. Extensive experiments demonstrate that significantly outperforms existing baselines,achieving state-of-the-art performance in both realism and diversity.
format Preprint
id arxiv_https___arxiv_org_abs_2605_02439
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Anomaly-Preference Image Generation
Wang, Fuyun
Wang, Yuanzhi
Guo, Xu
Huang, Sujia
Zhang, Tong
Wang, Dan
Yan, Hui
Liu, Xin
Cui, Zhen
Computer Vision and Pattern Recognition
Machine Learning
Synthesizing realistic and diverse anomalous samples from limited data is vital for robust model generalization. However, existing methods struggle to reconcile fidelity and diversity, often hampered by distribution misalignment and overfitting, respectively.To mitigate this, we introduce Anomaly Preference Optimization,a novel paradigm that reformulates anomaly generation as a preference learning problem.Central to our approach is an implicit preference alignment mechanism that leverages real anomalies as positive references, deriving optimization signals directly from denoising trajectory deviations without requiring costly human annotation. Furthermore, we propose a Time-Aware Capacity Allocation module that dynamically distributes model capacity along the diffusion timeline,prioritizing structural diversity during highnoise phases while enhancing fine-grained fidelity in low-noise stages. During inference, a hierarchical sampling strategy modulates the coherencealignment trade-off, enabling precise control over generation. Extensive experiments demonstrate that significantly outperforms existing baselines,achieving state-of-the-art performance in both realism and diversity.
title Anomaly-Preference Image Generation
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2605.02439