Lesson 3 advice for facial landmark detection

I tried to load the facial landmarks dataset used also in the pytorch tutorials (https://pytorch.org/tutorials/beginner/data_loading_tutorial.html). The loading finally went well somehow the training does not work well. Do you have any suggestion what I am doing wrong?
Here is the link to the kaggle notebook:
I turned off transformations, since when I tried with get_transforms() I could not create the databunch without a warning. I guess that cropping the image removed some landmarks and then the number of landmarks was not the same for all images…
Kind regards,