Getting 4 instead of 3 channels for image - Custom Dataset issue for CNN

I have been trying to implement plant seedling classification problem from Kaggle.While implementing it sometimes custom dataset function returning image shape as 4*224*224 instead of 3*224*224 after applying torchvision.transformations.

This is my colab notebook link :
Somebody help me with this issue.Thanks

I tried checking it, can you please link your dataset’s drive download link as well?
Maybe try checking the shape of val_dl if size mismatch is the case, it will reflect here.

Thanks for the response.This is the link for kaggle network.