arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2409.00768cs.CV
收藏

从训练数据视角重新思考图像超分辨率

Rethinking Image Super-Resolution from Training Data Perspectives

National Institute of Advanced Industrial Science and Technology (AIST)(日本产业技术综合研究所(AIST));Keio University(庆应义塾大学);University of Tsukuba(筑波大学);University of Amsterdam(阿姆斯特丹大学);University of Oxford(牛津大学);Tokyo Institute of Technology(东京工业大学)

Go Ohtani, Ryu Tadokoro, Ryosuke Yamada, Yuki M. Asano, Iro Laina, Christian Rupprecht, Nakamasa Inoue, Rio Yokota, Hirokatsu Kataoka, Yoshimitsu Aoki

AI总结:

本文从训练数据多样性和质量角度重新审视图像超分辨率,提出自动化图像评估流水线,发现低压缩伪影、高图像内多样性及大规模数据均能提升SR性能。

摘要:

在这项工作中,我们研究了图像超分辨率(SR)所用训练数据这一未被充分探讨的影响因素。通常,新的SR方法大多在DIV2K和DF2K等常见训练数据集上开发和评测。然而,我们从多样性和质量的角度对训练数据进行考察和重新思考,从而回应“SR训练对SR模型有多重要?”这一问题。为此,我们提出了一种自动化图像评估流水线。借助该流水线,我们对现有的高分辨率图像数据集以及ImageNet和PASS等更大规模图像数据集进行分层,以比较其性能。我们发现,具有以下特征的数据集均能正向提升SR性能:(i)低压缩伪影,(ii)以不同物体数量衡量的高图像内多样性,以及(iii)来自ImageNet或PASS的大量图像。我们希望所提出的简单而有效的数据集整理流水线能够为未来SR数据集的构建提供参考,并带来整体更优的模型。

英文摘要:

In this work, we investigate the understudied effect of the training data used for image super-resolution (SR). Most commonly, novel SR methods are developed and benchmarked on common training datasets such as DIV2K and DF2K. However, we investigate and rethink the training data from the perspectives of diversity and quality, {thereby addressing the question of ``How important is SR training for SR models?''}. To this end, we propose an automated image evaluation pipeline. With this, we stratify existing high-resolution image datasets and larger-scale image datasets such as ImageNet and PASS to compare their performances. We find that datasets with (i) low compression artifacts, (ii) high within-image diversity as judged by the number of different objects, and (iii) a large number of images from ImageNet or PASS all positively affect SR performance. We hope that the proposed simple-yet-effective dataset curation pipeline will inform the construction of SR datasets in the future and yield overall better models.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑