TF Graph that splits images and feeds the slices through model?

Is it possible to create a graph that splits an image into 4 pieces, then feeds each piece through a convolutional model one image at a time?

maybe with tf.image.extract_patches or tf.slice? and tf.function, and some if statements?

I want to do it all on-gpu / in-graph. 1 model.

I think that you can use tf.image.extract_patches.

You can check the ops device placement with: