Tree-based Semantic Losses: Application to Sparsely-supervised Large Multi-class Hyperspectral Segmentation

TL;DR

This paper introduces tree-based semantic loss functions for hyperspectral image segmentation, leveraging hierarchical label structures to improve performance with sparse annotations and enable out-of-distribution detection.

Contribution

The authors propose novel tree-based semantic loss functions that utilize hierarchical label information, enhancing segmentation accuracy with sparse labels and OOD detection in hyperspectral imaging.

Findings

Achieved state-of-the-art results on a 107-class hyperspectral dataset.

Effectively detects out-of-distribution pixels without harming in-distribution segmentation.

Demonstrated benefits of hierarchical label structure in semantic loss functions.

Abstract

Hyperspectral imaging (HSI) shows great promise for surgical applications, offering detailed insights into biological tissue differences beyond what the naked eye can perceive. Refined labelling efforts are underway to train vision systems to distinguish large numbers of subtly varying classes. However, commonly used learning methods for biomedical segmentation tasks penalise all errors equivalently and thus fail to exploit any inter-class semantics in the label space. In this work, we introduce two tree-based semantic loss functions which take advantage of a hierarchical organisation of the labels. We further incorporate our losses in a recently proposed approach for training with sparse, background-free annotations. Extensive experiments demonstrate that our proposed method reaches state-of-the-art performance on a sparsely annotated HSI dataset comprising $107$ classes organised in a clinically-defined semantic tree structure. Furthermore, our method enables effective detection of out-of-distribution (OOD) pixels without compromising segmentation performance on in-distribution (ID) pixels.