Han
Han

Reputation: 311

Keras CIFAR10 finetuning on VGG16: How can I preprocess the input data to fit the VGG16 network?

I was working on fine-tuning examples (currently on VGG16). I want to train the CIFAR10 dataset with VGG16 but the expected input dimensions are greater than 48x48px (CIFAR10 has 32x32).

I could not find a way to resize the image to fit in the network.

Please help me out!

Upvotes: 0

Views: 1185

Answers (1)

petezurich
petezurich

Reputation: 10184

You can simply set the input_shape to a dimension of your choosing.

Be aware that you very likely will get inferior results since VGG16 expects at least 48x48px. Quoting from Keras documentation:

"… width and height should be no smaller than 48. E.g. (200, 200, 3) would be one valid value."

Another way is to use flow_from_directory that gives you the option of resizing your images to any dimension you want:

https://keras.io/preprocessing/image/

Upvotes: 1

Related Questions