Convolutional Neural Network to classify MNIST images using Keras. It includes a training pipeline and an inference server.
Run inference server using docker. This starts the inference server on port 5000:
docker-compose up mnist
Do a POST call to the server root uploading an MNIST test image to infer its
class. The following calls the API with the test image located at
test/test_image.png:
curl -F 'image=@test/test_image.png' http://localhost:5000
Images need to be grayscale and 28px x 28px in size.
To train a new model the training pipeline can be used:
docker-compose run mnist /code/train.py --epochs 10 --batch-size 64
There is a prototype notebook located here