BriefGPT.xyz
Aug, 2024
损伤检测艺术中的最先进技术失败
State-of-the-Art Fails in the Art of Damage Detection
HTML
PDF
Daniela Ivanova, Marco Aversa, Paul Henderson, John Williamson
TL;DR
本研究解决了在传统媒介（如绘画、摄影、纺织品等）中准确检测和分类损伤的挑战，尤其在无先验知识的情况下机器学习模型的不足。提出的DamBench数据集包含超过11,000个标注，涵盖15种损伤类型，为损伤检测提供了新方法。研究发现当前模型在不同媒介类型之间泛化能力有限，强调了可靠损伤检测的必要性。
Abstract
Accurately detecting and classifying damage in
Analogue Media
such as paintings, photographs, textiles, mosaics, and frescoes is essential for cultural heritage preservation. While
Machine Learning
models excel i
→