search

yilmazkorkmaz1 / SSDiffRecon

Official TensorFlow implementation of Self-Supervised MRI Reconstruction with Unrolled Diffusion Models
MIT License
25 stars 0 forks source link

A question about validation #7

Open 9tchaser opened 1 month ago

9tchaser commented 1 month ago

Hi I read in your paper that you're using 10 subjects for validation, but I didn't see the correspoding code with validation process. May I ask how you implemented this part?

yilmazkorkmaz1 commented 1 month ago

Hi, sorry for late response. Unfortunately, I do not have clean version of that. We used corresponding PSNR and SSIM values in validation set to select network checkpoint, number of diffusion iterations (both forward and backward) and also self-supervised mask ratio. Self-supervised loss is also a good indicator for PSNR and SSIM so you can validate on the loss value directly.