chapter 2 has unknown mistake, please help me about beginners-pytorch-deep-learning HOT 6 OPEN

Hey, it's an PIL.UnidentifiedImageError – the image cannot be opened and identified.

1. Are you able to open the image ./train/cat\1004525_cba96ba3c3.jpg on your working station? Is it a real picture/valid file? Edit: The backslash after cat seems a bit weird to me: the images with the cats should be in an own directory called cat, I think it needs to be a slash instead of the backslash.
2. How have you acquired the image files? Have you used the download.py file or the gdrive link?

from beginners-pytorch-deep-learning.

Hey, it's an PIL.UnidentifiedImageError – the image cannot be opened and identified.

1. Are you able to open the image ./train/cat\1004525_cba96ba3c3.jpg on your working station? Is it a real picture/valid file? Edit: The backslash after cat seems a bit weird to me: the images with the cats should be in an own directory called cat, I think it needs to be a slash instead of the backslash.
2. How have you acquired the image files? Have you used the download.py file or the gdrive link?

_
thanks for your help!

I download this image package through G Drive, after your reminder, i check the package on my PC and found '1004525_cba96ba3c3.jpg' has broken, does it means i must use check_image()?

from beginners-pytorch-deep-learning.

Yes, please adjust your code for the three respective lines and add , is_valid_file=check_image. The check_image() function is included to avoid these kind of error messages and the broken images not being considered.

For example for the training data use the ImageFolder class as follows (see https://github.com/falloutdurham/beginners-pytorch-deep-learning/blob/master/chapter2/Chapter%202.ipynb):

train_data = torchvision.datasets.ImageFolder(root=train_data_path,
                                              transform=img_transforms,
                                              is_valid_file=check_image)

The same for val_data and test_data.

Is there any reason you deleted it in the code you forked from the gihub repo?

from beginners-pytorch-deep-learning.

If you got an FileNotFoundError with the check_image() function please check out #71.

from beginners-pytorch-deep-learning.

Yes, please adjust your code for the three respective lines and add , is_valid_file=check_image. The check_image() function is included to avoid these kind of error messages and the broken images not being considered.

For example for the training data use the ImageFolder class as follows (see https://github.com/falloutdurham/beginners-pytorch-deep-learning/blob/master/chapter2/Chapter%202.ipynb):

train_data = torchvision.datasets.ImageFolder(root=train_data_path,
                                              transform=img_transforms,
                                              is_valid_file=check_image)

The same for val_data and test_data.

Is there any reason you deleted it in the code you forked from the gihub repo?

Thanks for your patient!
The reason I delete 'img_check()' is I noticed the bug in torchvision you mentioned in another issue, it seems that the function is important for both download by Url and GDrive. Or we can update a new image package to GDrive.

from beginners-pytorch-deep-learning.

Yes, unfortunately also some GDrive images seem to be affected, but there is no issue at all as long it's possible to include the check_image() function.

I think it's best to downgrade torchvision to 0.10.1 as mentionend in #71 - or you can upgrade to 0.12.0 (released 2 days ago; the bug mentioned in #71 has been fixed in the release - but be aware that the new release has not been tested for this book repo yet).

from beginners-pytorch-deep-learning.