Figure 2 | Scientific Reports

Figure 2

From: Semantic modeling of cell damage prediction: a machine learning approach at human-level performance in dermatology

Figure 2

Qualitative examples from the segmentation model highlighting different aspects: (a) in case of low contrast, the performance drops considerably. (b) shows robustness to missing annotations (considered as label noise). Figure (c) shows a case where the metric (macro-average IoU) drops considerably (although qualitatively almost perfect) due to small predictions for absent classes (in this case for healthy epidermis (green) although completely damaged (red)).

Back to article page