AOS - Image Restoration

How to prepare the data

Create a new directory (e.g. aos-data). Within this directory create sub-folders which hold the different data versions (e.g. focalstack-1). In each directory which holds the version of the data, you need to create two further sub-folders train and test and split the data accordingly. In case you have a different structure like in the example below, you need to provide the path to the root-data folder (e.g. aos-data) as an argument (--data_path) to the main.py when running the script. The default values of the script expect the folder structure to be like the following:

|-- <project root>     
    |-- data
        |-- aos-data
            |--- focalstack-1
                |-- train
                    <1st focal plane image>.png
                    ...
                    <nth focal plane image>.png
                    <ground truth image>.png
                    ...
                
                |-- test
                    <1st focal plane image>.png
                    ...
                    <nth focal plane image>.png
                    <ground truth image>.png
            ...
                    
    |-- model
    ...

How to perform training or evaluation

Run python main.py in the root directory. An example for starting the training of a new model using the folder structure described above would be python main.py --train True --data_name focalstack-1. In order to see all the possible configurations run python main.py -h

How to see the results of the training

Run tensorboard --logdir result in the root directory.