When I was reproducing the results for unrolling length = 1–7, at t=1 it should be a pure CNN network applied to FastMRI. Under a 10% sampling rate, the PSNR is 30.98; then at t=2 the PSNR is 31.31, and at t=5 it is 31.75. The RFB module doesn’t seem to have much effect, and the influence of the number of iterations seems a bit odd. How did you implement this?
When I was reproducing the results for unrolling length = 1–7, at t=1 it should be a pure CNN network applied to FastMRI. Under a 10% sampling rate, the PSNR is 30.98; then at t=2 the PSNR is 31.31, and at t=5 it is 31.75. The RFB module doesn’t seem to have much effect, and the influence of the number of iterations seems a bit odd. How did you implement this?