search

11510880 / CG_VD_VAMP

CG_VD_VAMP for MRI-based Image construction
1 stars 0 forks source link

MSE calculations and image estimate #1

Open charlesmillard opened 4 years ago

charlesmillard commented 4 years ago

Hi,

Nice to see some work on VDAMP! Do you have a paper or dissertation or something you could share to explain the new methods?

When the MSE of VDAMP is retrospectively calculated, it uses hist.C_thr, the thresholded signal estimate. In your VDAMP_O code, you have renamed hist.C_thr to hist.c_1k_hat but not updated its name in the MSE calculation, which is why it returns the same x_hat and MSE estimate at every iteration - hist.C_thr is always just zeros. I have run your examples with this bug fix and VDAMP ends up with a similar MSE to your methods.

Thanks.

11510880 commented 4 years ago

Hi,

Thank you very much for pointing out the problem and thank you for your code which helped me a lot with my dissertation. I am afraid I am not yet allowed to share my dissertation before my viva. But the basic idea of my method is simple: replacing the back projection part of AMP with the LMMSE part of VAMP and do that per wavelet band as in VDAMP. And since there is a matrix inverse in the LMMSE, conjugate gradient was used to solve the inverse problem, for which you could refer to https://ieeexplore.ieee.org/document/9053799.