Fig. 9
From: Investigating the Quality of DermaMNIST and Fitzpatrick17k Dermatological Image Datasets
Visualizing the distributions of duplicates in Fitzpatrick17k, filtered by different combinations of criteria. Total counts are inset in each plot. \({{\mathscr{S}}}_{0.90}\) and \({{\mathscr{S}}}_{0.95}\) denotes pairs with similarity scores of at least 0.90 and 0.95 respectively. \(\widehat{{\mathscr{D}}}\) denotes pairs that differ in their diagnoses labels. \({\widehat{{\mathscr{F}}}}^{\ge 1}\) and \({\widehat{{\mathscr{F}}}}^{ > 1}\) denotes pairs that ‘differ in their FST labels by 1’ versus ‘by more than 1’, respectively. Best viewed online.
