python - How to write a custom Keras validation to average predictions over multiple image tiles before computing the validation metric

Question

I am currently training a deep learning model on images tiles (299x299) using Keras. Those tiles are extracted from larger images (4000x4000) and each tile inherits from his original image label. In the validation step, instead of computing the accuracy (or any other metric) based on each tiles prediction, I would like first to average the prediction of every tile from the same original image and then compute the accuracy over the batch size, but I don't see how to do it with Keras, any help would be appreciated :) I already made a custom validation generator, but i don't know how to first tell Keras to average prediction before computing the validation metric

Thanks!

Answer 1

等待大神答复