This repository provides an efficient representation for assessment of model calibration in machine learning / deep learning.
In particular, it allows to collect information about a segmentation model's confidence in every pixel or voxel of a number of cases and then allows to investigate model calibration per label or per case via a number of different measures.
Details can be found in our paper:
Farina Kock, Felix Thielke, Grzegorz Chlebus, and Hans Meine: Confidence Histograms for Model Reliability Analysis and Temperature Calibration Proc. Medical Imaging with Deep Learning 2022 https://openreview.net/forum?id=p2f6ROn1h02
If you find this code useful, we would be happy about feedback, contributions or citations of the above paper.