Does this skill support 3D point clouds?

Yes, it supports point cloud formats like PCD, PLY, LAS, and LAZ, and can automatically handle conversion and grouping for 3D visualization.

Do I need to specify the label format?

No, the skill performs a deep scan of the target directory to infer label formats (such as COCO, YOLO, or VOC) based on folder structure and file contents.

Can it handle multimodal data?

Absolutely. It detects patterns like scene folders or filename prefixes to group disparate media types like images, videos, and LiDAR into single FiftyOne samples.

Is it safe to use on existing FiftyOne datasets?

Yes, the skill explicitly checks for existing dataset names and asks whether you want to overwrite, rename, or abort before any data is modified.

What happens if a required package like 'pandaset' is missing?

The skill automatically detects the specific requirements for your data, checks your environment, and prompts for permission to install the necessary SDKs.

FiftyOne Dataset Import

Name: FiftyOne Dataset Import
Author: voxel51

byvoxel51

•

データサイエンスとML

Automates the detection and ingestion of computer vision datasets into FiftyOne with support for multimodal data and complex label formats.

概要

This skill streamlines the process of importing visual data into FiftyOne by automatically identifying media types, label formats, and directory structures. It handles everything from standard image and video formats to complex autonomous driving datasets like nuScenes or Waymo, manages required Python package installations, and correctly organizes multimodal grouped datasets. By performing deep directory scans and offering interactive validation, it eliminates the need for manual ingestion scripts and ensures data integrity for machine learning workflows.

主な機能

Automatic detection of COCO, YOLO, VOC, KITTI, and other standard formats
Support for multimodal grouped datasets including images and point clouds
Deep directory scanning to infer relationships and metadata structure
Automated identification and installation of required SDKs and packages
Interactive confirmation workflow to prevent dataset name collisions
3 GitHub stars

ユースケース

Visualizing complex autonomous driving datasets with 3D point clouds and cuboids
Rapidly importing local image folders with mixed annotation formats for model debugging
Standardizing messy data directories into structured FiftyOne datasets without writing code

概要

主な機能

Automatic detection of COCO, YOLO, VOC, KITTI, and other standard formats
Support for multimodal grouped datasets including images and point clouds
Deep directory scanning to infer relationships and metadata structure
Automated identification and installation of required SDKs and packages
Interactive confirmation workflow to prevent dataset name collisions
3 GitHub stars

ユースケース

Visualizing complex autonomous driving datasets with 3D point clouds and cuboids
Rapidly importing local image folders with mixed annotation formats for model debugging
Standardizing messy data directories into structured FiftyOne datasets without writing code