Skip to main content

Table 9 The performance of ARS on continuous space vs. AutoAugment on discrete space [116]

From: A survey on Image Data Augmentation for Deep Learning

Model AutoAugment ARS-Aug
Wide-ResNet-28-10 2.68 2.33
Shake-Shake (26 2 × 32 days) 2.47 2.14
Shake-Shake (26 2 × 96 days) 1.99 1.68
Shake-Shake (26 2 × 112 days) 1.89 1.59
AmoebaNet-B (6,128) 1.75 1.49
PyramidNet + ShakeDrop 1.48 1.26