To run:

python -W ignore olivetti.py --write-to-file True --generate-data True

The test-set accuracy should be close to the 97%.

We have 400 samples; 260 training images and 140 test images.

If we train the model on 260 images the maximum accuracy will be 5%.

python -W ignore olivetti.py --write-to-file True

We will have two types of image generator: train and test

train_data_gen = ImageDataGenerator(rescale=1./255, shear_range=0.2, 
                                    zoom_range=0.2, horizontal_flip=True)

So what does it mean?

From a selected sample:

10 samples are generated based on the shearing, zoom, and horizontal flip:

Each training samples are generated for 230 images

Each test samples are generated for 71 images.

Total 55.721 training images, 9935 test images.

Pull requests are welcome.

For major changes, please open an issue, then discuss what you would like to change.

MIT