jottenlips / guitar-classifier

🎸 Machine Learning for Guitars: Retraining InceptionV3 to classify guitars and Converting Tensorflow to CoreML for iOS (presentation code)

Guitar-Classifier

Machine Learning for Guitars - this is an elaboration on tensorflow for poets

Tensorflow

pip install virtualenv

virtualenv env

source env/bin/activate
pip install -tensorflow
pip install -pillow
pip install -tfcoreml

IMAGE_SIZE=224 ARCHITECTURE="mobilenet_0.50_${IMAGE_SIZE}"

Download images

I used https://github.com/hardikvasa/google-images-download CLI

Resize Images

change the paths in resize.py

cd scripts python resize.py

Retrain Model on your images

python -m scripts.retrain \
  --bottleneck_dir=tf_files/bottlenecks \
  --how_many_training_steps=500 \
  --model_dir=tf_files/models/ \
  --summaries_dir=tf_files/training_summaries/"${ARCHITECTURE}" \
  --output_graph=tf_files/retrained_graph.pb \
  --output_labels=tf_files/retrained_labels.txt \
  --architecture="${ARCHITECTURE}" \
  --image_dir=tf_files/Guitars

Test tensorflow model

python -m scripts.label_image \
    --graph=tf_files/retrained_graph.pb  \
    --image=../Desktop/sg.jpg

tensorboard --logdir tf_files/training_summaries &

Conversion to CoreML

cd scripts

Change all paths to where you want to read/write your model and test your images.

Test conversion

python labelcoreml.py

Convert model

python converter.py

From Core ML in ArKit, my fork from the presentation will be up soon. https://github.com/hanleyweng/CoreML-in-ARKit

https://github.com/jottenlips/GuitarClassifier-iOS My fork of that repo.