Each tutorial below is a curated demonstration of how FiftyOne can help refine your datasets and turn your good models into great models.
pandas-style queries in FiftyOne
Translate your pandas knowledge to FiftyOne. This tutorial gives a side-by-side comparison of performing common opertaions in pandas and FiftyOne.
Evaluating object detections
Aggregate statistics aren't sufficient for object detection. This tutorial shows how to use FiftyOne to perform powerful evaluation workflows on your detector.
Evaluating a classifier
Evaluation made easy. This tutorial walks through an end-to-end example of fine-tuning a classifier and understanding its failure modes using FiftyOne.
Using image embeddings
Visualize your data in new ways. This tutorial shows how to use FiftyOne's powerful embeddings visualization capabilities to improve your image datasets.
Annotating with CVAT
So you've loaded and explored your data in FiftyOne... but now what? See how to send it off to CVAT for annotation in just one line of code.
Annotating with Labelbox
Unlock the power of the Labelbox platform. See how you can get your FiftyOne datasets annotated with just one line of code.
Training with Detectron2
Put your FiftyOne datasets to work and learn how to train and evaluate Detectron2 models directly on your data.
Downloading and evaluating Open Images
Expand your data lake and evaluate your object detection models with Google's Open Images dataset and evaluation protocol, all natively within FiftyOne.
Exploring image uniqueness
Your models need diverse data. This tutorial shows how FiftyOne can remove near-duplicate images and recommend unique samples for model training.
Finding classification mistakes
Better models start with better data. This tutorial shows how FiftyOne can automatically find label mistakes in your classification datasets.
Finding detection mistakes
How good are your ground truth objects? Use the FiftyOne Brain's mistakenness feature to find annotation errors in your object detections.
Nearest Neighbor Embeddings Classification with Qdrant
Easily pre-annotate your FiftyOne datasets using approximate nearest neighbors search on embeddings with Qdrant.
Fine-tuning YOLOv8 model predictions
Visualize and evaluate YOLOv8 model predictions before fine-tuning for your custom use case.
Build 3D point cloud datasets with Point-E
Lidar is expensive. This tutorial shows how FiftyOne can help you construct high quality 3D point cloud datasets using Point-E point cloud models.
Check out the fiftyone-examples repository for more examples of using FiftyOne!