Explainable image segmentation with wavelet-network

Lieb, Hanna-Georgina and Kaszta, Tamás (2025) Explainable image segmentation with wavelet-network. In: Proceedings of the International Conference on Formal Methods and Foundations of Artificial Intelligence. Eszterházy Károly Katolikus Egyetem Líceum Kiadó, Eger, pp. 148-160. ISBN 9789634963035

Preview

Text fmfai2025_pp148-160.pdf - Published Version Download (2MB) | Preview

Abstract

Recent advances in artificial intelligence and its widespread adoption have imposed the necessity of research targeting the inner mechanisms of intelligent systems. We lack the exact mathematical tools needed to grasp what led to a certain output. The term explainability has recently emerged in the context of artificial intelligence (AI) as an area of development. An efficient way to introduce a checkpoint into decision-making systems is to incorporate prototype units. These bridge the difference between input image space and feature space, offering us a glimpse into an intermediary phase of decision-making. We created a new model – WaveProtoSeg – from the WaveProtoPNet classification model, combining image segmentation with the wavelet transform as a feature extractor. Our experiments were conducted on the Cityscapes dataset, which gathers real street scenes. Although we did not achieve the accuracy of the original paper, we explored various configurations of the system, and we managed to build a versatile system.

Actions (login required)

Edit Item