Where is the tutorial of 3D reconstruction with pairwise alignment?

[waniwal]

The tutorial of 3D reconstruction with pairwise alignment is not found.

[YifanLu2000]

Hi, thank you very much for your interest! We will update the tutorial as soon as possible to provide a smoother, more detailed, and comprehensive guide for users. However, as I have been quite busy these days and would like to make the tutorial as perfect as possible, there has been some delay. We expect to release the related tutorial this week and hope you can understand.

[yuling999666]

Hi @YifanLu2000, I would like to ask how can I reproduce the result of pairwise alignment since you have multiple randomness generator in your st.align.morph_align () function. I tried to set seed out of the function but I cannot generate the same result. Could you give some suggestions? Thanks.

[YifanLu2000]

Hi @yuling999666, I believe there are two sources of randomness in st.align.morph_align: one is the SVI process, and the other is the selection of inducing points, which are generated by torch.roll (when using GPU) and np.random.choice, respectively. Could you try setting the seed for both PyTorch and NumPy to see if that resolves the issue?

[yuling999666]

Hi @YifanLu2000 , thank you for your reply. It seems that you have used np.random.choice() multiple times in the morpho_pairwise(). If I just set seed before applying st.align.morph_align(), the first np.random.choice() will use my seed, but the second, the third one... will not. I think it's not decent to copy your code and change it inside the package. I have stuck in it for a long time... still figuring out how to do it.

[YifanLu2000]

Hi @yuling999666, after my attempts, I found that setting the random seed for PyTorch and NumPy before the function can ensure consistent results for each run. However, it’s important to note that this needs to be done alongside the execution of st.align.morpho_align. You can add the following two lines before calling st.align.morpho_align:

np.random.seed(42)
torch.manual_seed(42)
aligned_slices, _= st.align.morpho_align(...)

[yuling999666]

Hi @YifanLu2000 , thanks for reply. I appreciate it. I can generate the same results using the same conda environment, but when I change to another conda environment, it changes. I'm still trying to figure out if it's due to computer or server not the code itself.