JeremyGibson / gvision

Play around with the Google vision API

Photo classifier

Just a fun script to try out the Google Vision API. Thanks to @jmgibson71 for the code and inspiration. Just a note that using this API costs money. I currently have free credits that expire soon, but @jmgibson71 noted that it costs about $30 to label about 15,000 photos.

google setup

1. Login to your Google cloud account and create credentials for Vision API, which will download a credentials.json file to your computer. Copy that to this directory:

   cp ~/Downloads/vkurup-ckad-5033a83c8926.json my-vision-creds.json
   

2. Add a reference to that file to your environment variables:

   export GOOGLE_APPLICATION_CREDENTIALS="my-vision-creds.json"
   

3. Install the gcloud tool

4. Confirm that your authentication works:

   gcloud auth application-default print-access-token
   

setup

1. Create a virtualenv

2. Set your PHOTOS_DIR env var where this app will look for your photos (can be nested in folders):

   export PHOTOS_DIR=/home/vkurup/Pictures
   

3. Install requirements:

   pip install -U pip wheel
   pip install -Ur requirements.txt
   

4. Run the tool:

   python label_maker.py
   

   It will print out status on the console and you can find logs about only the photos it thinks are potentially documents in the logs directory.