Does it support exporting specific subsets of data?

Absolutely. You can choose to export the entire dataset, your current filtered view, or only specific selected samples.

What happens if the export directory already exists?

The skill includes a safety directive to check for existing directories and will prompt the user to confirm an overwrite or provide a different path.

Which formats are supported for detection datasets?

The skill supports industry standards including COCO, YOLOv4, YOLOv5, Pascal VOC, KITTI, CVAT Image, and TensorFlow Object Detection.

Can I export labels without copying the image files?

Yes, you can use the 'LABELS_ONLY' export type to generate annotation files (like COCO JSON) without duplicating your media assets.

FiftyOne Dataset Export

Name: FiftyOne Dataset Export
Author: voxel51

byvoxel51

•

Data Science & ML

Exports FiftyOne computer vision datasets to industry-standard formats like COCO, YOLO, and VOC for machine learning training.

About

This skill streamlines the complex process of converting FiftyOne datasets into standardized formats required for model training and data sharing. It enables users to export media and labels (or labels only) for detections, classifications, and segmentations while ensuring compatibility with common frameworks like YOLOv5 and COCO. By providing a guided workflow that includes dataset validation, absolute path management, and parameter discovery, it significantly reduces manual errors in computer vision data pipelines and preprocessing tasks.

Key Features

Flexible export modes for Media and Labels, Labels Only, or Media Only
Support for multiple formats including COCO, YOLOv5, VOC, CVAT, and CSV
Dynamic parameter discovery using FiftyOne operator schemas
Automated safety checks to prevent accidental directory overwriting
Ability to export entire datasets, filtered views, or selected samples
3 GitHub stars

Use Cases

Creating portable CSV or GeoJSON archives of dataset metadata for spreadsheet analysis
Backing up entire FiftyOne datasets including all metadata, brain runs, and evaluations
Preparing computer vision datasets for training YOLOv8 or COCO-based object detection models

About

Key Features

Flexible export modes for Media and Labels, Labels Only, or Media Only
Support for multiple formats including COCO, YOLOv5, VOC, CVAT, and CSV
Dynamic parameter discovery using FiftyOne operator schemas
Automated safety checks to prevent accidental directory overwriting
Ability to export entire datasets, filtered views, or selected samples
3 GitHub stars

Use Cases

Creating portable CSV or GeoJSON archives of dataset metadata for spreadsheet analysis
Backing up entire FiftyOne datasets including all metadata, brain runs, and evaluations
Preparing computer vision datasets for training YOLOv8 or COCO-based object detection models