Open zhaosongyi opened 3 years ago
The problem here is that torchvision.datasets.ImageFolder scans the directory provided assuming it contains subdirectories mapping to classes, containing images pertaining to those closes.
At some point, someone re-uploaded the celeba.zip
generated from a folder called img_align_celeba/
, so unzipping creates a folder with this name instead of putting the images into the celeba/
folder.
The next cell fails because it sees two folders, celeba/
and img_align_celeba/
, treats these as distinct classes for prediction, and complains that the celeba
class folder is empty of images.
Quick fix for anyone seeing this: just add these two lines at the bottom of the DOWNLOAD IMAGES
cell or in a new cell, but before the LOAD IMAGES
cell.
os.rmdir('./data/celeba/')
os.rename('./data/img_align_celeba', './data/celeba')
Better fix for the notebook: no need to create the celeba
directory in the first place, the folder renaming is still worth doing.
it has showed: Downloaded & Extracted https://www.dropbox.com/s/kgqofw91l8m1mk6/celeba.zip?dl=1 but excute the next cell : ... FileNotFoundError: Found no valid file for the classes celeba. Supported extensions are: .jpg, .jpeg, .png, .ppm, .bmp, .pgm, .tif, .tiff, .webp