5.4-visualizing-what-convnets-learn: VGG16 images are square

[MadIceTea]

Hi Mr Chollet. Our professor has been having a wonderful time teaching with your book, and we're reading through up to this chapter for this semester's course.

As I was looking over your ipynb against my code, I realised that while all the programming syntax (e.g. ln 17) are OK, the first part of ln16, in the first instance a dimension is used, the VGG16 model describes itself as trained on 224 x 244 (expected 224 x 224).

It is admittedly listed in the errata of the print edition book (https://manning-content.s3.amazonaws.com/download/9/972b1ae-dfc3-47c1-80d0-afa55c4b9415/Chollet_DeepLearninginPython_err3.html), but I see no reason why it shouldn't be fixed outright on GitHub. I tried a fix, which can be implemented if #90 is merged.

Thanks.