دورية أكاديمية
Uncertainty-aware Cross-Entropy for Semantic Segmentation
العنوان: | Uncertainty-aware Cross-Entropy for Semantic Segmentation |
---|---|
المؤلفون: | S. Landgraf, M. Hillemann, K. Wursthorn, M. Ulrich |
المصدر: | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol X-2-2024, Pp 129-136 (2024) |
بيانات النشر: | Copernicus Publications, 2024. |
سنة النشر: | 2024 |
المجموعة: | LCC:Technology LCC:Engineering (General). Civil engineering (General) LCC:Applied optics. Photonics |
مصطلحات موضوعية: | Technology, Engineering (General). Civil engineering (General), TA1-2040, Applied optics. Photonics, TA1501-1820 |
الوصف: | Deep neural networks have shown exceptional performance in various tasks, but their lack of robustness, reliability, and tendency to be overconfident pose challenges for their deployment in safety-critical applications like autonomous driving. In this regard, quantifying the uncertainty inherent to a model’s prediction is a promising endeavour to address these shortcomings. In this work, we present a novel Uncertainty-aware Cross-Entropy loss (U-CE) that incorporates dynamic predictive uncertainties into the training process by pixel-wise weighting of the well-known cross-entropy loss (CE). Through extensive experimentation, we demonstrate the superiority of U-CE over regular CE training on two benchmark datasets, Cityscapes and ACDC, using two common backbone architectures, ResNet-18 and ResNet-101. With U-CE, we manage to train models that not only improve their segmentation performance but also provide meaningful uncertainties after training. Consequently, we contribute to the development of more robust and reliable segmentation models, ultimately advancing the state-of-the-art in safety-critical applications and beyond. |
نوع الوثيقة: | article |
وصف الملف: | electronic resource |
اللغة: | English |
تدمد: | 2194-9042 2194-9050 |
Relation: | https://isprs-annals.copernicus.org/articles/X-2-2024/129/2024/isprs-annals-X-2-2024-129-2024.pdf; https://doaj.org/toc/2194-9042; https://doaj.org/toc/2194-9050 |
DOI: | 10.5194/isprs-annals-X-2-2024-129-2024 |
URL الوصول: | https://doaj.org/article/b312ff03484a4971b8ee3013cce30229 |
رقم الأكسشن: | edsdoj.b312ff03484a4971b8ee3013cce30229 |
قاعدة البيانات: | Directory of Open Access Journals |
تدمد: | 21949042 21949050 |
---|---|
DOI: | 10.5194/isprs-annals-X-2-2024-129-2024 |