Error in Google Colaboratory - AttributeError: module 'PIL.Image' has no attribute 'register_decoder'

Question

I am running this code on Google Colaboratory and I am getting error of register decoder

image_data = dset.ImageFolder(root="drive/SemanticDataset/train/", transform = transforms.Compose([
                                        transforms.Scale(size=img_size),
                                        transforms.CenterCrop(size=(img_size,img_size*2)),
                                        transforms.ToTensor(),
                                        ]))
enter code herelabel_data = dset.ImageFolder(root="drive/SemanticDataset/label/", transform = transforms.Compose([
                                        transforms.Scale(size=img_size),
                                        transforms.CenterCrop(size=(img_size,img_size*2)),
                                        transforms.ToTensor(),
                                        ]))

image_batch = data.DataLoader(image_data, batch_size=batch_size, shuffle=False, num_workers=2)
label_batch = data.DataLoader(label_data, batch_size=batch_size, shuffle=False, num_workers=2)
for i in range(epoch):
   for _, (image, label) in enumerate(zip(image_batch, label_batch)):
       optimizer.zero_grad()

       x = Variable(image, requires_grad=True).cuda()
       y = Variable(label).cuda()

       out = model.forward(x)
       loss = loss_func(out, y)

       loss.backward()
       optimizer.step()

       if _ % 100 == 0:
           print("Epoch: "+i+"| Loss: " , loss)

here is the screenshot of error

Answer 1

First, check the version of pillow you have by using:

import PIL
print(PIL.PILLOW_VERSION)

and make sure you have the newest version, the one I am using right now is 5.3.0 If you have like 4.0.0, install a new version by using: !pip install Pillow==5.3.0 in the Colab environment.

Second, restart your Google colab environment, and check the version again, it should be updated.

I had the same problem, and I spent some time trying to solve it.

Note: I was using PyTorch 0.4.

I hope this will solve your problem.