FiftyOne Examples

FiftyOne is an open source ML tool created by Voxel51 that helps you build high-quality datasets and computer vision models. You can check out the main github repository for the project here.

This repository contains examples of using FiftyOne to accomplish various common tasks.

Usage

Each example in this repository is provided as a Jupyter Notebook. The table of contents below provides handy links for each example:

  Click this link to run the notebook in Google Colab (no setup required!)

  Click this link to view the notebook in Jupyter nbviewer

  Click this link to download the notebook

Running examples locally

You can always clone this repository:

git clone https://github.com/voxel51/fiftyone-examples

and run any example locally. Make sure you have Jupyter installed and then run:

jupyter notebook examples/an_awesome_example.ipynb

Table of contents

Shortcuts	Examples	Description
	quickstart	A quickstart example for getting your feet wet with FiftyOne
	walkthrough	A more in-depth alternative to the quickstart that covers the basics of FiftyOne
	comparing_YOLO_and_EfficientDet	Compares the YOLOv4 and EfficientDet object detection models on the COCO dataset
	digging_into_coco	A simple example of how to find mistakes in your detection datasets
	deepfakes_in_politics	Evaluating deepfakes using a deepfake detection algorithm and visualizing the results in FiftyOne
	emotion_recognition_presidential_debate	Analyzing the 2020 US Presidential Debates using an emotion recognition model
	image_uniqueness	Using FiftyOne's image uniqueness method to analyze and extract insights from unlabeled datasets
	structured_noise_injection	Visually exploring a method for structured noise injection in GANs from CVPR 2020
	visym_pip_175k	Exploring the People in Public 175K Dataset from Visym Labs with FiftyOne
	wrangling_datasets	Using FiftyOne to load, manipulate, and export datasets in common formats
	open_images_evaluation	Evaluating the quality of the ground truth annotations of the Open Images Dataset with FiftyOne
	working_with_feature_points	A simple example of computing feature points for images and visualizing them in FiftyOne
	image_deduplication	Find and remove duplicate images in your image datasets with FiftyOne
	hardness_for_image_classification	Use the FiftyOne Brain to mine the hardest images in your classification dataset
	pytorch_detection_training	Using FiftyOne datasets to train a PyTorch object detection model
	pytorchvideo_model_evaluation	Evaluate and visualize PyTorchVideo models with FiftyOne
	training_clearml_detector	Train a model with ClearML and FiftyOne to detect DRAGONS!
	converting_tags_to_classifications	Convert classifications to tags and back to annotate them right in the FiftyOne App

Contributing

This repository is open source and community contributions are welcome!

Check out the contribution guide to learn how to get involved.

Citation

If you use FiftyOne in your research, feel free to cite the project (but only if you love it 😊):

@article{moore2020fiftyone,
  title={FiftyOne},
  author={Moore, B. E. and Corso, J. J.},
  journal={GitHub. Note: https://github.com/voxel51/fiftyone},
  year={2020}
}

If you use a specific contributed example in this repository, please also cite the author directly (if one is specified).