README.md

Detecting label issues in an object detection dataset

This example demonstrates how to use cleanlab to detect label errors in an object detection dataset. We train a neural network with the Detectron2 library with a model from the COCO-Detection model zoo.

There are two notebooks:

Notebook	Description
detectron2_training.ipynb	Trains an object detection model on a training set of images and produces predictions on a held-out validation set.
detectron2_training-kfold.ipynb	Trains an object detection model on a training set of images via k-fold cross-validation to produce predictions on that same training set.

Setup

Before running the notebooks, make sure you install dependencies (ideally in a fresh virtual environment) with:

# # Set up a virtual environment
# python -m venv venv
# source venv/bin/activate

pip install -r requirements.txt

Also, make sure you have wget installed.