Cannot load NASNetLarge

[korytmar]

My code is:

m = NASNetLarge(include_top=False,weights='imagenet',input_shape=(224,224,3))

But this line of code gives me this error:

ValueError: When settinginclude_top=True and loading imagenet weights, input_shape should be (331, 331, 3).

It only works with shape (331,331,3) but for larger inputs like (640,640,3) does not work again.

[gabrieldemarmiesse]

This is normal. The NASNetLarge isn't made to accept different input sizes, like the VGG16 or VGG19 for example. The documentation states:

The default input size for the NASNetLarge model is 331x331 and for the NASNetMobile model is 224x224

[korytmar]

I know that, but this default values are also for other models.

In docs NASNetLarge:

'It should have exactly 3 inputs channels, and width and height should be no smaller than 32. E.g. (200, 200, 3) would be one valid value.'

According to this, I think it should work or am I misunderstanding?

[gabrieldemarmiesse]

I see what you mean. Sorry, I misread you the first time. I'm going to dig in, but I get the same error, and it's true that it makes no sense! Thanks for insisting!

[gabrieldemarmiesse]

You are right, I did a check in more depths. A PR is going to be made. It'll take quite some time to make it in the stable release.