bartorch.ssim#
- bartorch.ssim(reference, input)#
Mean structural similarity of the magnitudes (
--ssim).Magnitudes are the root sum of squares over the coil axis, and each image is divided by the maximum magnitude of its
reference. SSIM withk1 = 0.01,k2 = 0.03, dynamic range one and population statistics over a uniform 7x7 window in(y, x), at every position the window fits whole (no padding), averaged over those positions, overzand over every other axis (networks/losses.c:304-327,nn/losses.c:151-285).- Parameters:
reference (torch.Tensor) – C order
(..., coils, z, y, x); a plain(y, x)image serves.yandxare at least seven.input (torch.Tensor) – C order
(..., coils, z, y, x); a plain(y, x)image serves.yandxare at least seven.