Could you add a deploy function to the superresolution project?

[zhuofeng]

I have to test my own images with the superresolution functon, but I found it hard to deal with it because the pre-processing of image(average_pooling3d) transformed it into tensor. Could you write a deploy program of superresolution, or tell me how to write? Thank you.

[mrajchl]

Hi there,
generally these examples are designed to showcase an application (i.e. how to build the network, train, etc). The SR network here is very simplistic, meaning that it is likely not something you want to properly deploy anyways (c.f. dltk.networks.super_resolution.simple_super_resolution).

However, if you would still like to do that, you could refer to one of the other deploy scripts (e.g. regression) and adapt the script from there. It should be quite similar. You would have to run the inference in PREDICT mode (https://github.com/DLTK/DLTK/blob/master/examples/applications/IXI_HH_superresolution/train.py#L80) and supply a 4-fold downsampled image (https://github.com/DLTK/DLTK/blob/master/examples/applications/IXI_HH_superresolution/train.py#L29) from a 1mm resolution (c.f. https://github.com/DLTK/DLTK/blob/master/examples/applications/IXI_HH_superresolution/reader.py#L43).

We'll see if we get around to writing a deploy script for this application, but since this is just a showcase, it will be quite low priority on the list. If you solve the problem yourself, we'd be happy to take a pull request from you.

If you need more help and want to chat with one of us, please contact us on the gitter chat: https://gitter.im/DLTK/DLTK

HTH,
M