a) Raw EM data can be found in  /ccdbprod/ccdbprod23/home/CCDB_DATA_USER.portal/CCDB_DATA_USER/acquisition/project_20271/microscopy_5391395/segmentation/testAppOut/em

b) Membrane Prediction folder can be found in /ccdbprod/ccdbprod23/home/CCDB_DATA_USER.portal/CCDB_DATA_USER/acquisition/project_20271/microscopy_5391395/segmentation/testAppOut/predictions-003

c) Myelin Prediction folder can be found in /ccdbprod/ccdbprod23/home/CCDB_DATA_USER.portal/CCDB_DATA_USER/acquisition/project_20271/microscopy_5391395/segmentation/testAppOut/predictions-007

Step 1

Find the closest membrane model in the existing models. In this case, we want to use membranes_Aug05_40k, which can be downloaded at https://doi.org/10.7295/W9CDEEP3M50772

Step 1.1 Create the retrain data:

Go to membrane prediction data (a). Create the binary version of the prediction data by using the ImageMagick command:

convert $input.png -threshold 50% $output.png

Step 1.2 Create the augmented data using CDeep3M

PreprocessTrainingData.py /images /labels /mem_augmented

Step 1.3 Feed the retrain data to the model (membranes_Aug05_40k) that I picked

runtraining.sh --additerations 5000 --retrain ~/W9CDEEP3M50772 ~/mem_augmented ~/mem_retrain_out

Step 2

Step 2.1 Convert the Myelin prediction result (c) to the binary format:

convert $input.png -threshold 50% $output.png

Step 2.2 Create the augmented data using CDeep3M:

PreprocessTrainingData.py /myelin_images /myelin_labels /myelin_augmented  

Step 2.3  Create the training model:

runtraining.sh --numiterations 50000 /myelin_augmented  /myelin_trainednet

Step3