Loading request...
The user wants more flexible input management, allowing association of inputs into named tuples or concatenation into channels of the same tensor. This would give application writers more control over mapping inputs to readers.
NiftyNet could benefit from more flexible input management. At the moment, the config file allows a user to associate a set of inputs into a named input tuple as follows: ``` images = image_t1, image_t2 ``` The user can also concatenate a set of inputs into channels of the same tensor as follows: ``` images = [image_t1, image_t2] ``` This is as much configuration as the user has; the application writer is then responsible for mapping inputs to input readers and so forth. It would be beneficial to create more explicit input groups that can be used interchangeably with inputs and specify more flexible mappings between data sources and tensors. See the [Config File Refactors](https://github.com/NifTK/NiftyNet/wiki/Config-File-Refactors) page in the wiki for more details.