search

seannz / svr

[CVPR2024] Fully convolutional slice-to-volume reconstruction for single-stack MRI
https://seaniyoung.com/publications.html
MIT License
24 stars 6 forks source link

Training and weights for adult brain data #6

Open tomjf opened 1 month ago

tomjf commented 1 month ago

Could you provide some more information in the README on how data should be pre-processed and setup for training on adult brain. Also could you provide the pre-trained weights for the adult data as well?

Many thanks, Tom

seannz commented 4 weeks ago

Hello Tom,

Training on adult brains is not too different from training on fetal — in your brain3d_svr method call, you'd just need to point to a folder containing e.g. 256x256x256 volumes of pre-aligned (e.g. MNI-space) scans. Most datasets would already come with pre-aligned volumes.

I'll upload the pre-trained weights for the adult network to our google drive and ping you again — since publication, we have made some changes to the order of strided convs in our network model and there seems to be a mismatch between our weights and the (slightly updated) model definition.