SUCCESS will be in the form of something like this:
Test set: Average loss: 0.0300, Accuracy: 4/4 (100%)
You can train the model yourself by python mnist.py --train and you can switch off pretrained as well:
python mnist.py --train --no_pretrained
There is a test image in img/ with the name test_2.png so you can run:
python predict_interface_usage.py test_2 to predict the handwritten digits.
SUCCESS will write an output with predictions to the command
prompt and it will generate an image with the predictions in /pro-img/