Modzy - Label Studio Sample Project

📺 Follow along on YouTube: https://youtu.be/ZOxeNHzIbfg

This sample project demonstrates how to use Modzy and Label studio to dynamically re-label production inferences using a combination of semi-supervised learning, explainable AI, and human-in-the-loop review. This is a comprehensive and low-lift way to create training data sets from production data that can be used to retrain a machine learning model.

This sample project uses:

For model inference and explainability

As a makeshift data lake for storing images

For labeling production inference results

Installation & Set-Up

Label Studio Set-up

• First, install Label Studio and start it up. Instructions can be found here (installing with PIP was a breeze)
• Once it's running, log in and download your API access token
• Finally, create a project in Label Studio. Then go to "Settings" > "Labeling Interface" and paste in label-classes.html as code

Modzy Set-up

In your instance of Modzy, create a project and then download your project api key.

Dropbox Set-up

• If you don't have one, create a free Dropbox account
• Then, you'll need to create a Dropbox App
• Next, give your app the following permissions: files.metadata.write files.content.write files.content.read sharing.write
• Finally, generate an access token

Environment Set-Up

Before you can start running the sample images as inferences, you'll need to do the following:

• Clone this repo
• [Optional, but recommended] Create a virtual environment within your project folder and activate it
• Run $ pip install -r requirements.txt to install all necessary dependencies
• Create a copy of .env-sample and rename it to .env. Then update it to include your API access tokens for Modzy, Dropbox, and Label Studio
• Run $ source .env to load your environment variables
• Update the labelStudioURL variable in import-annotations.py to the URL of your instance of Label Studio

Usage

Send sample images to Modzy and Dropbox

The next step will send the images in the /images-test folder to a model that executes "Image-based Geolocation" which means that it will try to predict where in the world the image was taken. Then the image will be uploaded to Dropbox.

In your terminal run:

$ python3 inference.py

Run import-annotations.py

Finally, send all predictions generated from Modzy over to Label Studio for review and labeling.

In your terminal run:

$ python3 import-annotations.py