The concept is to remove noise from a black and white images. Such as :
The following command will train your neural network from scratch on 10 epochs. On my computer it takes around 40s each epoch :
glowingspoon --train=True --train_x="my/path/modified" --train_y="my/path/original/"
--save_nn="model.pth" --epochs=10
In order to test, you need to specify the testing flag and some images to test. For example :
glowingspoon --train=False --test=True
--test_x="/new/path/folder/test/modified/"
--test_y="/new/path/folder/test/original/"
--load_nn="model.pth"
--print_example=True
-
By running :
glowingspoon --help
you will recieve a lot of indication and the default values for each parameters.
-
The measurement is the structural similarity from scikit-learn. The mean value is about 0.80 over the whole images.
- cross-validation nor k-fold validation
- This isn't a perfect tool, there is a lot of improvement that can be done. It's not very stable and this code is mostly a proof of concept on a "research question".