Personalized Vision via Visual In-Context Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jiang, Yuxin, Gu, Yuchao, Song, Yiren, Tsang, Ivor, Shou, Mike Zheng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914064313090048
author Jiang, Yuxin
Gu, Yuchao
Song, Yiren
Tsang, Ivor
Shou, Mike Zheng
author_facet Jiang, Yuxin
Gu, Yuchao
Song, Yiren
Tsang, Ivor
Shou, Mike Zheng
contents Modern vision models, trained on large-scale annotated datasets, excel at predefined tasks but struggle with personalized vision -- tasks defined at test time by users with customized objects or novel objectives. Existing personalization approaches rely on costly fine-tuning or synthetic data pipelines, which are inflexible and restricted to fixed task formats. Visual in-context learning (ICL) offers a promising alternative, yet prior methods confine to narrow, in-domain tasks and fail to generalize to open-ended personalization. We introduce Personalized In-Context Operator (PICO), a simple four-panel framework that repurposes diffusion transformers as visual in-context learners. Given a single annotated exemplar, PICO infers the underlying transformation and applies it to new inputs without retraining. To enable this, we construct VisRel, a compact yet diverse tuning dataset, showing that task diversity, rather than scale, drives robust generalization. We further propose an attention-guided seed scorer that improves reliability via efficient inference scaling. Extensive experiments demonstrate that PICO (i) surpasses fine-tuning and synthetic-data baselines, (ii) flexibly adapts to novel user-defined tasks, and (iii) generalizes across both recognition and generation.
format Preprint
id arxiv_https___arxiv_org_abs_2509_25172
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Personalized Vision via Visual In-Context Learning
Jiang, Yuxin
Gu, Yuchao
Song, Yiren
Tsang, Ivor
Shou, Mike Zheng
Computer Vision and Pattern Recognition
Machine Learning
Modern vision models, trained on large-scale annotated datasets, excel at predefined tasks but struggle with personalized vision -- tasks defined at test time by users with customized objects or novel objectives. Existing personalization approaches rely on costly fine-tuning or synthetic data pipelines, which are inflexible and restricted to fixed task formats. Visual in-context learning (ICL) offers a promising alternative, yet prior methods confine to narrow, in-domain tasks and fail to generalize to open-ended personalization. We introduce Personalized In-Context Operator (PICO), a simple four-panel framework that repurposes diffusion transformers as visual in-context learners. Given a single annotated exemplar, PICO infers the underlying transformation and applies it to new inputs without retraining. To enable this, we construct VisRel, a compact yet diverse tuning dataset, showing that task diversity, rather than scale, drives robust generalization. We further propose an attention-guided seed scorer that improves reliability via efficient inference scaling. Extensive experiments demonstrate that PICO (i) surpasses fine-tuning and synthetic-data baselines, (ii) flexibly adapts to novel user-defined tasks, and (iii) generalizes across both recognition and generation.
title Personalized Vision via Visual In-Context Learning
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2509.25172