This repository contains code for a data augmentation method RICAP (Random Image Cropping And Patching) based on Data Augmentation using Random Image Cropping and Patching for Deep CNNs implemented in PyTorch.
WideResNet28-10 baseline on CIFAR-10:
python train.py --dataset cifar10WideResNet28-10 +RICAP on CIFAR-10:
python train.py --dataset cifar10 --ricap TrueWideResNet28-10 +Random Erasing on CIFAR-10:
python train.py --dataset cifar10 --random-erase TrueWideResNet28-10 +Mixup on CIFAR-10:
python train.py --dataset cifar10 --mixup True| Model | Error rate | Loss | Error rate (paper) |
|---|---|---|---|
| WideResNet28-10 baseline | 3.82 | 0.158 | 3.89 |
| WideResNet28-10 +RICAP | 2.82 | 0.141 | 2.85 |
| WideResNet28-10 +Random Erasing | 3.18 | 0.114 | 4.65 |
| WideResNet28-10 +Mixup | 3.02 | 0.158 | 3.02 |
Learning curves of loss and accuracy.
