Dear fenglinglwb,
First of all, thank you for your interesting package.
I would like to apply this package for medical images. In this domain, we have some images that are all of them cropped from the lower sides (20-40% of the lower part is missing). We want to complete this image using MAT.
I applied the CelebA-HQ_512.pkl model and I got the following results:
Ground Truth
Applied Mask
Result
Since in your training set, there were no medical images, I should train the network again. Therefore, I have two questions:
1- Do you think this method for our problem will be successful?
2- If it is possibly successful, training from scratch would be beneficial, or finetuning your existing pre-trained model?
Thank you for your time.
Best regards,
Javad
Dear fenglinglwb, First of all, thank you for your interesting package. I would like to apply this package for medical images. In this domain, we have some images that are all of them cropped from the lower sides (20-40% of the lower part is missing). We want to complete this image using MAT. I applied the CelebA-HQ_512.pkl model and I got the following results:
Ground Truth
Applied Mask Result
Since in your training set, there were no medical images, I should train the network again. Therefore, I have two questions: 1- Do you think this method for our problem will be successful? 2- If it is possibly successful, training from scratch would be beneficial, or finetuning your existing pre-trained model? Thank you for your time. Best regards, Javad