🌪️ DamageArbiter: A CLIP-Enhanced Multimodal Arbitration Framework for Hurricane Damage Assessment from Street-View Imagery

This repository contains the implementation, figures, and dataset links for the paper:
“DamageArbiter: A CLIP-Enhanced Multimodal Arbitration Framework for Hurricane Damage Assessment from Street-View Imagery.”

---

📘 Overview

This study proposes a Disagreement-driven Arbitration Framework designed to improve the interpretability, reliability, and accuracy of street-view-based disaster damage assessment.
It systematically combines Vision Transformer (ViT) and CLIP (Contrastive Language–Image Pretraining) representations, supported by LLM-generated disaster annotations.

---

🎥 Demo Video

Watch the demonstration of DamageArbiter on YouTube.

---

🧩 Methodology Framework

The framework integrates:

• Vision-based feature extraction (ViT)
• LLM-assisted textual annotation generation
• CLIP-based cross-modal alignment
• Confidence-based arbitration for explainable disaster damage prediction

---

📊 Figures

Figure 1. Study Area

Figure 2. Label Example

Figure 5. CLIP Model

Figure 8. Spatial Mapping Results

---

📂 Dataset

You can access the street-view disaster dataset from the following DOI:

Yang, Yifan (2025)
Perceiving Multidimensional Disaster Damages from Street–View Images Using Visual–Language Models
📁 figshare Dataset DOI: 10.6084/m9.figshare.28801208.v2

or

The primary hosting platform is Hugging Face Datasets, which provides a version-controlled repository for convenient access, inspection, and integration with machine learning workflows:

🔗 https://huggingface.co/datasets/Rayford295/BiTemporal-StreetView-Damage

The dataset includes:

• Pre- and post-disaster street-view imagery
• Georeferenced location and damage type annotations
• Severity levels (mild, moderate, severe)
• Sample image regions from Horseshoe Beach, Florida, after Hurricane Milton

---

🏛️ Conference Presentation (AAG 2026)

This work has been accepted for presentation at the 2026 Annual Meeting of the American Association of Geographers (AAG 2026) and received the 🥈2nd Place Award in the AAG GIS Specialty Group Student Honors Paper Competition.

• Session: AAG GIS Specialty Group — Honors Competition for Student Papers
• Presentation Type: Gallery Presentation (Student Honors Competition)
• Official AAG Link: https://aag-meetings.secure-platform.com/aag2026/gallery/rounds/149/details/90541

Presentation Schedule:

• Date: Tuesday, March 17, 2026
• Time: 4:10 PM – 5:30 PM
• Location: Imperial B, Ballroom Level, Hilton Union Square

---

📄 Preprint (arXiv)

Yang, Y., Zou, L., Gong, W., Fu, K., Li, Z., Wang, S., ... Tian, H. (2026).
DamageArbiter: A CLIP-Enhanced Multimodal Arbitration Framework for Hurricane Damage Assessment from Street-View Imagery.
arXiv preprint arXiv:2603.14837
🔗 https://arxiv.org/abs/2603.14837

📌 Citation

If you find this work useful, please consider citing:

@article{yang2026damagearbiter,
  title     = {DamageArbiter: A CLIP-Enhanced Multimodal Arbitration
               Framework for Hurricane Damage Assessment from
               Street-View Imagery},
  author    = {Yang, Yifan and Zou, Lei and Gong, Wenjing and
               Fu, Kai and Li, Zhen and Wang, Shuo and others
               and Tian, H.},
  journal   = {arXiv preprint arXiv:2603.14837},
  year      = {2026}
}

---

⚠️ Usage and Permissions

All codes, figures, and datasets in this repository were developed and curated solely for academic research purposes as part of
“ DamageArbiter: A Disagreement-driven Arbitration Framework for Hurricane Damage Assessment from Street-View Imagery.”

If you wish to reuse, reproduce, modify, or distribute any portion of the codebase, figures, or dataset, please contact the author in advance to obtain written permission.

📩 Contact:
Yifan Yang (yyang295@tamu.edu)
Department of Geography, Texas A&M University
🌐 https://rayford295.github.io

🚫 Unauthorized redistribution, adaptation, or commercial use of the materials in this repository is strictly prohibited.