Scholarlink

Structural damage assessment after natural disasters is a critical yet challenging task due to limited labeled data and the emergence of new damage patterns. In this work, we explore the use of vision-language models (VLMs) for post-disaster structural classification using the PeerHub ImageNet dataset. We evaluate several VLMs, including CLIP, OpenCLIP, SigLIP, and EvaCLIP, in a zero-shot setting across multiple structural tasks. Additionally, we investigate few-shot adaptation using Tip-Adapter and a simple prototype-based approach. Our results show that VLMs offer strong baseline performance, with few-shot methods improving reliability in low-data scenarios. To our knowledge, this is the first comprehensive study of VLMs for post-disaster structural assessment.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Zero-Shot and Few-Shot Learning with Vision-Language Models for Post-disaster Structural Damage Assessment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Oriol Chacón-Albero,
  • Jaume Jordán,
  • Vicent Botti,
  • Vicente Julian

摘要

Structural damage assessment after natural disasters is a critical yet challenging task due to limited labeled data and the emergence of new damage patterns. In this work, we explore the use of vision-language models (VLMs) for post-disaster structural classification using the PeerHub ImageNet dataset. We evaluate several VLMs, including CLIP, OpenCLIP, SigLIP, and EvaCLIP, in a zero-shot setting across multiple structural tasks. Additionally, we investigate few-shot adaptation using Tip-Adapter and a simple prototype-based approach. Our results show that VLMs offer strong baseline performance, with few-shot methods improving reliability in low-data scenarios. To our knowledge, this is the first comprehensive study of VLMs for post-disaster structural assessment.