I think the training dataset is not proper, there are images in the middle which are distorted, please let me know.
Dataset is not proper
Training data seems to have no problem, images can be of different size but that should not be a problem, just resize them while loading them. Images may be little distorted, like its part of data augmentation, but each of them will be rectangular only. Try to resize all of them in your model specific size, say 100*100. If still you face any problem please provide further information.
Hope this helped You.
image data before 330 images are fine but after that they all are distorted, please load the npy file and check the images after 330.
Yes data till 328 is correct only. After that there’s an error with the file. So use only first 328 images. Just confirmed it with the mentor.
Hope this solved your doubt. 