About the dilation

[leon0n]

hallo guys, thanks for your sharing. But I do have some questions
About the dilation operation, in your paper, you show the result the best result with dilation = 5, but I can't find any dilation operation in this project. Also I tried by myself, once I dilated the label mask image, the pixel value of defect which should be in range(0,1) would be changed to (0, <1 like 0.69), so should I do the binarization to the dilated label mask? or not ?

hope to get your response

[skokec]

Hi, the prepared dataset file for TensorFlow already has dilation=5 included. To get other dilations, you can download the original png images from here and then use this script to generate the TensorFlow dataset with the desired dilations.

Search for the _process_dataset() function for the entry point which takes dilation parameter. This function will also do the binarization of label when require_binary_output = True, which we did use in our case.

Best, Domen.

[leon0n]

Thanks for your response!
I have got a better result on the Dataset .
Thank you