search

IAmSuyogJadhav / 3d-mri-brain-tumor-segmentation-using-autoencoder-regularization

Keras implementation of the paper "3D MRI brain tumor segmentation using autoencoder regularization" by Myronenko A. (https://arxiv.org/abs/1810.11654).
MIT License
369 stars 113 forks source link

3 things have been changed. #30

Closed Crispy13 closed 4 years ago

Crispy13 commented 4 years ago
  1. VAE branch output will be included model output. (VAE branch will be trained)
  2. Minor change in dice score calculation (regardless of batch size)
  3. SpatialDropout3D will be used instead of Dropout.
IAmSuyogJadhav commented 4 years ago

Please confirm if you have tested these functions to work correctly. Are the model weights changing as expected after this change?

Crispy13 commented 4 years ago

Please confirm if you have tested these functions to work correctly. Are the model weights changing as expected after this change?

Yes. The VAE branch is contained in the model and now trained well.