Enregistré dans:
Détails bibliographiques
Auteurs principaux: Lee, Sanghyeon, Lee, Minwoo, Shin, Euijin, Kim, Kangyeol, Choi, Seunghwan, Choo, Jaegul
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2603.27637
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866908919502209024
author Lee, Sanghyeon
Lee, Minwoo
Shin, Euijin
Kim, Kangyeol
Choi, Seunghwan
Choo, Jaegul
author_facet Lee, Sanghyeon
Lee, Minwoo
Shin, Euijin
Kim, Kangyeol
Choi, Seunghwan
Choo, Jaegul
contents We introduce a parameter-efficient adaptation method for panel-aware in-context image generation with pre-trained diffusion transformers. The key idea is to compose learnable, panel-specific orthogonal operators onto the backbone's frozen positional encodings. This design provides two desirable properties: (1) isometry, which preserves the geometry of internal features, and (2) same-panel invariance, which maintains the model's pre-trained intra-panel synthesis behavior. Through controlled experiments, we demonstrate that the effectiveness of our adaptation method is not tied to a specific positional encoding design but generalizes across diverse positional encoding regimes. By enabling effective panel-relative conditioning, the proposed method consistently improves in-context image-based instructional editing pipelines, including state-of-the-art approaches.
format Preprint
id arxiv_https___arxiv_org_abs_2603_27637
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle OPRO: Orthogonal Panel-Relative Operators for Panel-Aware In-Context Image Generation
Lee, Sanghyeon
Lee, Minwoo
Shin, Euijin
Kim, Kangyeol
Choi, Seunghwan
Choo, Jaegul
Computer Vision and Pattern Recognition
We introduce a parameter-efficient adaptation method for panel-aware in-context image generation with pre-trained diffusion transformers. The key idea is to compose learnable, panel-specific orthogonal operators onto the backbone's frozen positional encodings. This design provides two desirable properties: (1) isometry, which preserves the geometry of internal features, and (2) same-panel invariance, which maintains the model's pre-trained intra-panel synthesis behavior. Through controlled experiments, we demonstrate that the effectiveness of our adaptation method is not tied to a specific positional encoding design but generalizes across diverse positional encoding regimes. By enabling effective panel-relative conditioning, the proposed method consistently improves in-context image-based instructional editing pipelines, including state-of-the-art approaches.
title OPRO: Orthogonal Panel-Relative Operators for Panel-Aware In-Context Image Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.27637