Data-Augmentation-Pytorch

Data-Augmentation example based on torchsample

1. Introduction.

This is an example which adopts torchsample package to implement data augmentation. This package provides many data augmentation methods such as rotation, zoom in or out.

2. Demo

The CIFAR-10 classification task is used to show how to utilize this package to implement data augmentation.

2.1. Run Demo:

python main.py

and

python advanced_main.py

2.2. Implementation:

Standard method: (random horizontal flip data augmentation.)

import torchvision.transforms as transforms
import torchsample as ts
train_tf= transforms.Compose([
            transforms.RandomHorizontalFlip(), # data augmentation: random horizontal flip
            transforms.Scale(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ])

Adding rotation data augmentation:

import torchvision.transforms as transforms
import torchsample as ts
train_tf= transforms.Compose([
            transforms.RandomHorizontalFlip(),
            transforms.Scale(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            ts.transforms.Rotate(20), # data augmentation: rotation 
            ts.transforms.Rotate(-20), # data augmentation: rotation
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ])

2.3. Accuracy.

The final accuracy is shown as follows:

Network	Baselines	Data Augmentation/(Rotation)
AlexNet (No pretrained)

2.4. Convergence Curve.

3. Acknowledgement.

The torchsample is a very awesome package implemented by Nick Cullen.

tianbaochou/Data-Augmentation-Pytorch