---
格式版本: 2
标题: "The Impact of CutMix on Reliability and Robustness in Semantic Segmentation"
原文链接: "https://arxiv.org/abs/2608.18715"
发布日期: "2026-08-19"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:local:strict_original_body"
发布时间证据: "**\\[v1\\]** Wed, 19 Aug 2026 09:16:12 UTC (2,384 KB)"
发布时间校准原因: "规则确认唯一严格发布时间，来源 local:strict_original_body"
发布时间校准置信度: "high"
发布时间候选数量: 18
发布时间严格候选数量: 6
发布时间原页读取状态: "source template page reused from URL open"
发布时间未找到原因: ""
发布时间校准时间: "2026-08-20T15:22:45+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-20T15:18:01+08:00"
入库时间: "2026-08-20T07:22:45.676Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "https://arxiv.org/search/?query=deployment&searchtype=all"
匹配关键词:
  - "deployment"
  - "performance"
  - "AI"
相关厂家:
  []
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 5
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "论文主题为CutMix对语义分割模型可靠性的影响，属于计算机视觉领域，与超节点/AI Rack等主题完全无关，仅因摘要含deployment一词误命中。"
AI质检模型: "ali-deepseek-v4-flash"
AI质检时间: "2026-08-20T15:26:41+08:00"
AI主题相关性: 0
AI来源权威性: 2
AI新颖性: 1
AI技术细节: 0
AI商业部署信号: 0
AI完整性: 2
AI摘要: "该研究系统评估了CutMix数据增强对语义分割模型可靠性的影响，分别测试CNN架构DeepLabV3+与Transformer架构SegFormer；"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:19:51.685Z"
采集批次: "2026年8月20日14点19分32秒"
采集批次ID: "20260820-141932-079"
去重键: "https://arxiv.org/abs/2608.18715"
---

## Computer Science > Computer Vision and Pattern Recognition

## Title:The Impact of CutMix on Reliability and Robustness in Semantic Segmentation

Authors:[Steven Landgraf](https://arxiv.org/search/cs?searchtype=author&query=Landgraf,+S), [Markus Ulrich](https://arxiv.org/search/cs?searchtype=author&query=Ulrich,+M)

[View PDF](https://arxiv.org/pdf/2608.18715) [HTML (experimental)](https://arxiv.org/html/2608.18715v1)

> Abstract:Ensuring not only high accuracy but also reliable and robust predictions is critical for the deployment of semantic segmentation models in safety-critical applications such as autonomous driving. Despite the widespread use of CutMix - a simple yet powerful data augmentation strategy - its effect on the reliability and robustness in dense predictions tasks remains unexplored. Motivated by recent findings that semi-supervised segmentation methods, where CutMix is a core component, can severely degrade reliability, this study isolates and systematically analyzes the influence of CutMix on segmentation accuracy, calibration, and uncertainty quality. We evaluate two representative architectures, the CNN-based DeepLabV3+ and the transformer-based SegFormer, across both in-domain and out-of-domain scenarios. Our results show that CutMix has only a minor impact on segmentation accuracy but consistently improves the reliability, particularly under distribution shifts. These improvements indicate that CutMix primarily enhances the trustworthiness of the model's calibration and uncertainty rather than the raw segmentation prediction itself. This distinction is crucial for safety-critical deployment, where reliable confidence estimates are as important as raw performance.

| Comments: |  |
| --- | --- |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | [arXiv:2608.18715](https://arxiv.org/abs/2608.18715) \[cs.CV\] |
|  | (or [arXiv:2608.18715v1](https://arxiv.org/abs/2608.18715v1) \[cs.CV\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.18715](https://doi.org/10.48550/arXiv.2608.18715) |

## Submission history

From: Steven Landgraf \[[view email](https://arxiv.org/show-email/47bc716e/2608.18715)\]  
**\[v1\]** Wed, 19 Aug 2026 09:16:12 UTC (2,384 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.18715) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))