Fig. 4
From: Multi-sample \(\zeta \)-mixup: richer, more realistic synthetic samples from a p-series interpolant
Visualizing the results obtained using mixup  (b) and \(\zeta \)-mixup  (c, d, e) on images (a) from the ISIC 2017 dataset, with three values of \(\gamma \ (2.4, 2.8, 4.0)\) used for \(\zeta \)-mixup. Note how mixup  synthesizes unrealistic images with ghosting (selected images highlighted in blue in b), as evidenced by either multiple lesions overlapping or with artifacts (dark corners, stickers, ink markers) overlapping the lesion. On the other hand, for all values of \(\gamma \), \(\zeta \)-mixup  produces visibly more realistic images