279: super-resolution based on Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network

[stian-johnsen]

Status

READY

Description

An implementation of the network mentioned in the title, and in ticket 279. Provides a means of upsampling tensors through a purely convolutional (sub-) network. Cannot be trained or used for inference with NiftyNet, directly, and is therefore only usable as a network building block.

Types of changes

• [ ] Bug fix (non-breaking change which fixes an issue)
• [x] New feature (non-breaking change which adds functionality)
• [ ] Breaking change (fix or feature that would cause existing functionality to change)
• [x] My code follows the code style of this project.
• [x] I have added tests to cover my changes.
• [x] All new and existing tests passed.

Impacted Areas in Application

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

• niftynet.layer (new layer class, does not break or otherwise interfere with anything).

[fbragman]

I know Ryutaro Tanno has the EPSCN network already implemented as part of his work on "Bayesian Image Quality Transfer with CNNs: Exploring Uncertainty in dMRI Super-Resolution". I can talk to him about and see if I can port his own TF implementation to NiftyNet.