BeerClassifier

Classify your beer bottle images using Core ML and Keras / Turi Create.
Demo app for this presentation: Core ML 🏃 iOS Engineer

Note that after I tap Download New Model, app starts recognizing Ho*gaarden as well.

Requirements

• Swift 4.2
• Xcode 10.0
• Python 3.6
• virtualenv
• pip

Setup

virtualenv -p python3 ./
source ./bin/activate
pip install -r requirements.txt

Usage

Before running any script, run:

source ./bin/activate

General scripts

scraper.py

Downloads images using Microsoft Cognitive API. Note that you need to register Azure and set your azure key as AZURE_KEY in .env.

python scripts/scraper.py

quantize.py

python scripts/quantize.py -i ./models/BeerClassifier.mlmodel -o ./models/BeerClassifierQuantized.mlmodel

Keras-targetted scripts

keras/preprocess.py

Crops and resize all images.

python scripts/keras/preprocess.py

keras/create_data_csv.py

Creates train_data.csv and test_data.csv.

python scripts/keras/create_data_csv.py

keras/train.py

Trains using collected images to create model file.

python scripts/keras/train.py

keras/hdf5_to_mlmodel.py

Converts keras model (in HDF5) file to .mlmodel format.

python scripts/keras/hdf5_to_mlmodel.py

TuriCreate-targetted scripts

turicreate/convert_image_format.py

Convert images to turicreate-supporting formats.

python scripts/turicreate/convert_image_format.py

turicreate/create_sframe.py

Create an sframe file from prepared images.

python scripts/turicreate/create_sframe.py

turicreate/train.py

Trains using prepared images and and sframe file to create model file.

python scripts/turicreate/train.py