deepinfer / slicer-deepinfer Goto Github PK

Add support for auxiliary inputs and outputs to the json model descriptor.

Hi,

is it normal that the extension is that listed anymore?
(not found in extension list)

thanks

Would like to submit a deep learning model -- submitted form on website, and e-mailed [email protected] directly. Is this project still active and supported?

Tried installing DeepInfer through the 3D Slicer (4.11.0-2019-07-15-linux-amd64) Extension Manager, and received the following error on launch:

  File "/home/mlp6/.config/NA-MIC/Extensions-28391/DeepInfer/lib/Slicer-4.11/qt-scripted-modules/DeepInfer.py", line 1, in <module>
    import Queue
ModuleNotFoundError: No module named 'Queue'
loadSourceAsModule - Failed to load file "/home/mlp6/.config/NA-MIC/Extensions-28391/DeepInfer/lib/Slicer-4.11/qt-scripted-modules/DeepInfer.py"  as module "DeepInfer" !
Fail to instantiate module  "DeepInfer"
The following modules failed to be instantiated:
   DeepInfer

• when downloading a docker image (right now the UI freezes for a while which makes it less attractive to use)
• if Docker daemon is not running

While tryting to run DeepInfer with CT Bone Segmenter under 3D Slicer 4.8.1 (build myself), I get the following error and the process stops :
sitk::ERROR: The file "LocalPath.deepinfer.tmp\OutputLabel.nrrd" does not exist.

Some models share the same docker image. For example, prostate-segmenter and prostate-needle-finder are both inside the deepinfer/prostate.
When a user downloads one of the models it should automatically put the other models into the local JSON folder and make them accessible through Slicer-DeepInfer.

Many operations performing regression or classification would produce table output. It would be particularly nice to have support for tools like pyradiomics using Docker (AIM-Harvard/pyradiomics#240) rather than manually building the extension (https://github.com/Radiomics/SlicerRadiomics)

In Slicer 4.9 with Qt 5, slicer crashes when the current local model is either prostate segmenter or prostate needle finder and we switch to one of either of these two.
this doesn't happen with other models or in Slicer 4.8

Currently only uint8 images are supported (https://github.com/DeepInfer/Slicer-DeepInfer/blob/master/DeepInfer/DeepInfer.py#L642) which is very limiting for techniques like CT. It would be good to have support for short and floating point images based on the json files

If docker-py is available, it could potentially simplify docker interaction from the module.

https://github.com/docker/docker-py