中文

基于 FFT 的统计量选择与优化用于严重损坏图像的鲁棒识别

计算机视觉与模式识别 2025-10-13 v2

摘要

在图像损坏的情况下提升模型鲁棒性,是在机器人代理等智能设备上实现鲁棒视觉系统的关键挑战之一。特别是,鲁棒的测试时性能对大多数应用至关重要。本文提出了一种新颖的方法,用于提升任何分类模型的鲁棒性,尤其是在严重损坏的图像上。我们的方法 (FROST) 利用高频特征来检测输入图像的损坏类型,并选择逐层的特征归一化统计量。FROST 为不同模型和数据集提供了最先进的结果,在 ImageNet-C 上比竞争对手的相对增益高达 37.1%,在严重损坏上将 40.9% mCE 的基线性能提升了。

关键词

引用

@article{arxiv.2403.14335,
  title  = {FFT-based Selection and Optimization of Statistics for Robust Recognition of Severely Corrupted Images},
  author = {Elena Camuffo and Umberto Michieli and Jijoong Moon and Daehyun Kim and Mete Ozay},
  journal= {arXiv preprint arXiv:2403.14335},
  year   = {2025}
}

备注

ICASSP 2024. Copyright 2024 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other