NifTK / NiftyNet

[unmaintained] An open-source convolutional neural networks platform for research in medical image analysis and image-guided therapy
http://niftynet.io
Apache License 2.0
1.36k stars 404 forks source link

Regression application amendment for output identity resizing #371

Closed pritesh-mehta closed 5 years ago

pritesh-mehta commented 5 years ago

Status

READY

Description

Updated regression application to:

  1. Prevent cropping operation if the loss_border is None or 0. This is necessary in the case of regressing a single value of shape 1x1x1, where the use of cropping layer will throw a shape error.
  2. Allow regression output to maintain label shape when the resize sampler is chosen, rather than resize the output back to image shape. Again, the motivation was the case of regressing a single value of shape 1x1x1, where it is desirable for the output to not be resized to image shape. This change involved adding a new config flag --force_output_identity_resizing and using this flag in the regression application to choose the identity aggregator.

Types of changes

Todos

Impacted Areas in Application

List general components of the application that this PR will affect: *