Repository navigation
pretrained VAE provided by MAISI doesn't work well #1915
Description
when i put 3D CT(.nii.gz) into the pretained VAE provided by MAISI, using autoencoder.reconstruct function, the reconstruct image looks different compared with the input one. it becomes wider
Can you tell me where the problem is
Very very appreciate!
Activity
paproaskwar commentedon Jan 3, 2025
right is original drawing, and the left is the reconstruct image. the reconstruct image looks much "fatter" than the original, but they have a same shape(256, 256, 256)
paproaskwar commentedon Jan 6, 2025
[-]pretrained VAE doesn't work well[/-][+]pretrained VAE provided by MAISI doesn't work well[/+]paproaskwar commentedon Jan 7, 2025
I konw why...it seems that the output has a default space(1mm, 1mm, 1mm), which is different from input(1.5, 1.5, 2). Now im trying to find out how to change the output's space
KumoLiu commentedon Jan 7, 2025
Please refer to https://github.com/Project-MONAI/tutorials/tree/main/generation/maisi#31-3d-autoencoder-training.