Adversarial Examples Improve Image Recognition

Benchmark Model Rank Results
domain-generalization-on-vizwizEfficientNet-B8 (advprop+autoaug)#4Accuracy - All Images: 50.5Accuracy - Corrupted Images: 45.8
domain-generalization-on-vizwizEfficientNet-B7 (advprop+autoaug)#5Accuracy - All Images: 49.7Accuracy - Corrupted Images: 45
domain-generalization-on-vizwizEfficientNet-B6 (advprop+autoaug)#6Accuracy - All Images: 49.6Accuracy - Corrupted Images: 44.7
domain-generalization-on-vizwizEfficientNet-B5 (advprop+autoaug)#7Accuracy - All Images: 49.1Accuracy - Corrupted Images: 44
domain-generalization-on-vizwizEfficientNet-B4 (advprop+autoaug)#10Accuracy - All Images: 48.1Accuracy - Corrupted Images: 42.5
domain-generalization-on-vizwizEfficientNet-B3 (advprop+autoaug)#15Accuracy - All Images: 45.5Accuracy - Corrupted Images: 39.8
domain-generalization-on-vizwizEfficientNet-B2 (advprop+autoaug)#19Accuracy - All Images: 44.3Accuracy - Corrupted Images: 38.2
domain-generalization-on-vizwizEfficientNet-B1 (advprop+autoaug)#23Accuracy - All Images: 42.4Accuracy - Corrupted Images: 36.2
domain-generalization-on-vizwizEfficientNet-B0 (advprop+autoaug)#34Accuracy - All Images: 40.5Accuracy - Corrupted Images: 34.2
image-classification-on-imagenetAdvProp (EfficientNet-B8)#213Top 1 Accuracy: 85.5%Number of params: 88M
image-classification-on-imagenetAdvProp (EfficientNet-B7)#240Top 1 Accuracy: 85.2%Number of params: 66M