Hello, I hope you're doing well. I'm reaching out to ask about the best metrics for evaluating unpaired 3D image translation tasks. Specifically, I'm interested in using SSIM, PSNR, and FID to assess the quality of images translated from CT to MRI.
Since the images are unpaired, with no exact correspondence between CT and MRI slices, comparing them directly poses a challenge. The translated images often blend content from the source domain (CT) with the style of the target domain (MRI), further complicating evaluation.
Could you guide me on how to apply these metrics—SSIM, PSNR, and FID—in a standardized and meaningful way for this type of task?