Why transform to 224 x 224

[pjfalbe]

In the tutorial you transform the images to 224 x 224 why is that? Is because the model was trained with 224x224 images the modal may not be able to label larger images?

[zmughal]

Yes, that is the specific input size that the model used in the example takes and how it was trained.

Each model has something called a signature that tells you the dimensions and type it takes for input and gives for output. Currently I recommend using the saved_model_cli script to help work that out. The specifics of how the input and output are encoded into the TFTensor varies from model to model, so you have to read the model documentation to find out what is expected.