MAISI Diffusion model training

[cbe135]

For MAISI Diffusion model training, if dataset has image of varying size, thus need to use batch size of 1 for training. Does it make more sense to train with batch size 1? Or would it be better to resize all images to same size. And if so what would be the ideal size to use? Thank you.

[dongyang0122]

If the batch size is larger than 1, the case where a mini-batch contains patches of different sizes requires additional handling, which may not be trivial.

[cbe135]

Understood, I would keep it as batch 1 for the time being. Thank you.