MAISI VAE training crops

[MTGGIT]

Hi! Thanks so much for the great work on MAISI model.
According to the paper, the VAE is trained on random cropped patches of size [64,64,64] and later [128,128,128]. Could you please clarify how latent features are handled when performing inference / training of LDM on larger full-volume inputs? Looking forward to your insights!

[Can-Zhao]

Large volumes are processed by sliding window inference.