Imperial College London
Portrait Picture

Dr Ben Glocker

Faculty of Engineering, Department of Computing

Professor in Machine Learning for Imaging
 
 
 
//

Contact

 

+44 (0)20 7594 8334b.glocker Website CV

 
 
//

Location

 

377Huxley BuildingSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@article{Li:2023:10.1109/TMI.2023.3242838,
author = {Li, Z and Kamnitsas, K and Ouyang, C and Chen, C and Glocker, B},
doi = {10.1109/TMI.2023.3242838},
journal = {IEEE Transactions on Medical Imaging},
pages = {1885--1896},
title = {Context label learning: improving background class representations in semantic segmentation},
url = {http://dx.doi.org/10.1109/TMI.2023.3242838},
volume = {42},
year = {2023}
}
Download

RIS format (EndNote, RefMan)

TY  - JOUR
AB  - Background samples provide key contextual information for segmenting regionsof interest (ROIs). However, they always cover a diverse set of structures,causing difficulties for the segmentation model to learn good decisionboundaries with high sensitivity and precision. The issue concerns the highlyheterogeneous nature of the background class, resulting in multi-modaldistributions. Empirically, we find that neural networks trained withheterogeneous background struggle to map the corresponding contextual samplesto compact clusters in feature space. As a result, the distribution overbackground logit activations may shift across the decision boundary, leading tosystematic over-segmentation across different datasets and tasks. In thisstudy, we propose context label learning (CoLab) to improve the contextrepresentations by decomposing the background class into several subclasses.Specifically, we train an auxiliary network as a task generator, along with theprimary segmentation model, to automatically generate context labels thatpositively affect the ROI segmentation accuracy. Extensive experiments areconducted on several challenging segmentation tasks and datasets. The resultsdemonstrate that CoLab can guide the segmentation model to map the logits ofbackground samples away from the decision boundary, resulting in significantlyimproved segmentation accuracy. Code is available.
AU  - Li,Z
AU  - Kamnitsas,K
AU  - Ouyang,C
AU  - Chen,C
AU  - Glocker,B
DO  - 10.1109/TMI.2023.3242838
EP  - 1896
PY  - 2023///
SN  - 0278-0062
SP  - 1885
TI  - Context label learning: improving background class representations in semantic segmentation
T2  - IEEE Transactions on Medical Imaging
UR  - http://dx.doi.org/10.1109/TMI.2023.3242838
UR  - http://arxiv.org/abs/2212.08423v1
UR  - https://ieeexplore.ieee.org/document/10038608
UR  - http://hdl.handle.net/10044/1/102889
VL  - 42
ER  -
Download