Confusion about the `ref` variable in `Aseq_DIP_Sparse_view_CT.py`

[batman47steam]

Aseq_DIP/Aseq_DIP_Sparse_view_CT.py

Line 114 in f2e5074

ref = Variable(torch.rand((1,1,width,length)).cuda(), requires_grad=True)

Aseq_DIP/Aseq_DIP_Sparse_view_CT.py

Line 108 in f2e5074

optimizer = optim.Adam(net.parameters(), lr = learning_rate)

Aseq_DIP/Aseq_DIP_Sparse_view_CT.py

Line 118 in f2e5074

optimizer2 = optim.Adam([ref], lr = 1e-1)

Aseq_DIP/Aseq_DIP_Sparse_view_CT.py

Line 157 in f2e5074

optimizer.step()

Thank you for sharing this impressive work! I truly appreciate the effort put into it. However, I have some questions about certain details in the code. Specifically, I noticed that (\lambda) is initialized as a trainable random variable and is optimized by Optimizer 2. That said, I couldn’t find updates for Optimizer 2 in the later stages, nor did I observe the current output being used as the input for the next iteration. Looking forward to your reply. Thank you!

[ZijianGu0430]

I have the same question with batman47steam,why optimizer2 not be used?And the current output not be used as the input for the next iteration in ct reconstruction.