Apply trained model when gpu is unavailable

[gschlafly]

Description

Neural network deconvolution fails when the device does not have, or recognize, a GPU.

Files

• \src\napari_lf\lfa\neural_nets\LFNeuralNetworkProto.py
• src\napari_lf_widgetLF.py

  To reproduce

  Create a virtual environment with the following

  conda create -n napari-test3 -c conda-forge python=3.9 -y
  conda activate napari-test3
  python -m pip install "napari[all]"
  git clone --single-branch --branch NN-reduced https://github.com/PolarizedLightFieldMicroscopy/napari-LF.git
  cd napari-LF
  pip install -e .
  pip install torch
  pip install pytorch_lightning
  pip install torchvision
  pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

  then open napari. Use the trained NN model in the mouse brain dataset to deconvolve a light field image.

Screenshots

Solution ideas

• Alter the way the model is loaded, as described here. A solution to allow the deconvolution to be successful on the workstation, would be to make the workstation return True for torch.cuda.is_available.

[gschlafly]

Fixed with #46