pytorch imgaug

from imgaug import augmenters as iaa
from PIL import Image
import numpy as np
import torch
from torchvision import transforms,datasets
class BLUR(object):

    def __init__(self, param=0.5):
        self.param=0.5
        self.seq=iaa.Sequential([iaa.Fliplr(1.0)])

    def __call__(self, img):

        img=np.array(img)
        img=self.seq.augment_images([img])
        img=Image.fromarray((img[0]))
        return img
    

    def __repr__(self):
        return self.__class__.__name__ + '()'
topil=transforms.ToPILImage()
tsfm=transforms.Compose([
                transforms.Resize((232, 232)),  # force resize
                #transforms.RandomCrop(224),
                #transforms.RandomHorizontalFlip(0.05),
                #transforms.RandomGrayscale(0.02),
                #transforms.RandomRotation(10),
                transforms.ColorJitter(0.05, 0.05, 0.05, 0.05),
                 BLUR(),
                transforms.ToTensor()])
image_datasets=datasets.ImageFolder('/home/test',tsfm)
dataloader=torch.utils.data.DataLoader(image_datasets,num_workers=2,batch_size=4)
index=0
for batch_idx, data in enumerate(dataloader):
    inputs, labels = data
    for i in range(4):
        topil(inputs[i]).save('/home/output/'+str(index)+'.jpg')
        index+=1