FiftyOne Dataset Inference

概要

The FiftyOne Dataset Inference skill streamlines the transition from raw local media to structured machine learning datasets. It allows users to automatically scan directories, import images, videos, and point clouds, and handle labels in formats like COCO, YOLO, and VOC. Beyond data management, it provides a guided workflow for running model inference from the FiftyOne Model Zoo—including popular architectures like YOLOv8 and ResNet—enabling rapid visualization and comparison of ground truth labels against model predictions within the FiftyOne App.

主な機能

• Support for COCO, YOLO, VOC, KITTI, and CVAT dataset imports
• One-click inference for detection, classification, and segmentation models
• Seamless integration with the FiftyOne Model Zoo and App visualization
• End-to-end dataset management from creation to persistent storage
• 0 GitHub stars
• Automated directory scanning for media and label format detection

ユースケース

• Loading raw image directories to run object detection and generate initial predictions
• Importing pre-labeled COCO or YOLO datasets to validate annotations and model performance
• Converting folder-based classification trees into searchable, queryable FiftyOne datasets