- Keywords: Metric Learning, Digital Pathology, Patch Classification
- TL;DR: A simple yet effective way to boost patch classification in Digital Pathology via Metric Learning.
- Abstract: We consider the problem of patch classification in digital pathology. We introduce a simple yet effective way to boost patch classification performance via metric learning. We hypothesize that the self perturbation and contrastive loss are both useful in improving generalization of the classification model. In our experiments with the PCam dataset, we showed that models trained with both losses indeed outperformed our baseline where only cross-entropy loss is used. In addition, we also achieved state-of-the-art results on the PCam dataset.
- Code of conduct: I have read and accept the code of conduct.
- Remove if rejected: Remove submission from public view if paper is rejected.