Random within __getitem__?

vision
1

Hi,
I am trying to add white noise to images via data loader. However, the seed is constant - meaning same seed for the whole run. How to change the seed every epoch for example?

def __getitem__(self,index):
    img2read = self.imgs_name[index]
    img = Image.open(img2read)
    img = img.convert('L')
    if self.transform is not None:
        img = self.transform(img)
    label = img
    if (self.noise_std==-1):
        noise_std = randint(0,55)/255.0
    else:
        noise_std = self.noise_std
    img = img+noise_std*torch.randn(img.size()).type(torch.FloatTensor)
    return img,label
2

With numpy, changing the seed from above with np.random.seed() works. Any other, more elegant way?