search

ge-xing / Diff-UNet

Diff-UNet: A Diffusion Embedded Network for Volumetric Segmentation. (using diffusion for 3D medical image segmentation)
Apache License 2.0
149 stars 21 forks source link

Question on BTCV #11

Open qb02su opened 1 year ago

qb02su commented 1 year ago

  1. self.embed_model = BasicUNetEncoder(3, 1, 2, [64, 64, 128, 256, 512, 64]) self.model = BasicUNetDe(3, 14, 13, [64, 64, 128, 256, 512, 64]

why the embed_model not be BasicUNetEncoder(3, 1, 13, [64, 64, 128, 256, 512, 64]) 2. why use the resample_img(), and if I run on my own dataset, in which case I should use it.

920232796 commented 1 year ago
  1. The third parameter of the embed_model is useless, because embed_model is an encoder which only extract the features of image.
  2. 3D images need to be resampled to the same spacing, and you also need to resize the model's outputs to the original size in the testing phase.