KIT | KIT-Bibliothek | Impressum | Datenschutz

Loss Scheduling for Class-Imbalanced Segmentation Problems

Taubert, Oskar; Götz, Markus; Schug, Alexander; Streit, Achim

Abstract (englisch):
When training a classifier the choice of loss function heavily influences the characteristics of the resulting model.
The most commonly used loss function for classification is cross entropy. In image segmentation problems where each pixel is assigned to a particular class, overlap-based losses have recently been shown to improve classifier performance especially for datasets with an imbalanced class distribution. This is particularly relevant to segmentation because class imbalance mitigation strategies used in regular classification are often not applicable. Overlap-based losses, however, have different drawbacks. We are aiming at combining the upsides of different losses with a simple scheduling scheme during training while minimizing their downsides. Gradually transitioning from an overlap-based dice loss to cross entropy allows to reliably select a distinct minimum in the optimization landscape as a valuable alternative to results obtained from traditional unscheduled loss functions. We demonstrate the efficacy of our approach on different combinations of loss functions, datasets, and models.



Originalveröffentlichung
DOI: 10.1109/ICMLA51294.2020.00073
Zugehörige Institution(en) am KIT Steinbuch Centre for Computing (SCC)
Publikationstyp Proceedingsbeitrag
Publikationsjahr 2020
Sprache Englisch
Identifikator ISBN: 978-1-72818-470-8
KITopen-ID: 1000127713
HGF-Programm 46.12.02 (POF III, LK 01) Data Activities
Erschienen in 19th IEEE International Conference on Machine Learning and Applications (ICMLA)
Veranstaltung 19th International Conference on Machine Learning and Applications (2020), Online, 14.12.2020 – 17.12.2020
Verlag Institute of Electrical and Electronics Engineers (IEEE)
Seiten 422-427
Schlagwörter Machine Learning, Loss Scheduling, Semantic Segmentation, Pixel Classification, Model Calibration, Class Imbalance, Transfer Learning, Cross Entropy, Dice Loss
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page