search

Luchixiang / EMDiffuse

MIT License
25 stars 1 forks source link

Reproduce vEMDiffuse-i on EPFL brain dataset (Supp. Fig.18) #4

Closed raynehe closed 14 hours ago

raynehe commented 6 days ago

Hi! Thanks for your wonderful work!

I'm trying to run EMDiffuse-i on EPFL brain dataset using vEMDiffuse-i.json (as shown in Supp. Fig.18). I'm a little confused about how to prepare the data for isotropic reconstruction.

I found data_root in vEMDiffuse-i.json which is ended with /train_img, but the EPFL dataset is simply one .tif file. Should I preprocess the data the same way as denoising and super-resolution (using register.py)?

Specifically, how can I reproduce the results in Supp. Fig.18 (which parameters should I set)?

Sorry for the bother and looking forward to your reply!

Luchixiang commented 4 days ago

Hi. As I mentioned in the README. You should split the 3D .tif file into 2D tif file: 0.tif, 1.tif, 2.tif, and then run vEMDiffuse-i. The register file is used for the registration of 2D super-resolution and denoise.

As for the parameter, just use the default parameter and train the model for ~1000 epochs.

raynehe commented 14 hours ago

Thanks for your help!