I use Keras to create a convolutional neural network to perform regression from microscopic images to 2D tag data (for counting). I study the network in small areas of microscopic data (where patches are the size of the susceptible field). The problem is that the method fit()requires that the validation data be the same size as the input. Instead, I hope that I can check all the images (not patches) so that I can check my entire set of checks and compare the results with other methods that I have used so far.
One solution that I found was to alternate between each fit()and evaluate()every era. However, I was hoping I could observe these results with the Tensorboard. Since it evaluate()does not accept callbacks, this solution is not ideal. Does anyone have a good way to check full resolution images during corrections training?
source
share