Hello, thanks for your great work!
I’ve noticed a potential issue with the FMDD results reported in Table 2 of your paper. For FMDD and FastMRI, the normalize in line 291 of ttt_mim_online.py is not implemented. However, while instancing PSNR in line 435, it seems that you only take fastMRI into consideration for denormalize. Therefore, the PSNR results of FMDD may not be true. For example, the average PSNR for raw 48 images is 27.22 as reported in `A Poisson-Gaussian Denoising Dataset with Real Fluorescence Microscopy Images', which is far less than 39.10 for average 10 raw images in your paper.
A possible solution would be to update line 435 in ttt_mim_online.py as follows:
metric = {'psnr': PSNR().cuda() if args.dataset != 'fastmri' and args.dataset != 'fmdd' else PSNR(denormalize=False).cuda()}
This adjustment should ensure consistent behavior across datasets.
Thanks again for your great contributions! I hope you’ll consider this fix.
Hello, thanks for your great work!
I’ve noticed a potential issue with the FMDD results reported in Table 2 of your paper. For FMDD and FastMRI, the
normalizein line 291 ofttt_mim_online.pyis not implemented. However, while instancing PSNR in line 435, it seems that you only take fastMRI into consideration for denormalize. Therefore, the PSNR results of FMDD may not be true. For example, the average PSNR for raw 48 images is 27.22 as reported in `A Poisson-Gaussian Denoising Dataset with Real Fluorescence Microscopy Images', which is far less than 39.10 for average 10 raw images in your paper.A possible solution would be to update line 435 in
ttt_mim_online.pyas follows:This adjustment should ensure consistent behavior across datasets.
Thanks again for your great contributions! I hope you’ll consider this fix.