# Import Vision Datasets into FiftyOne

Dataset imports often fail when media, annotations, and multimodal folders use different formats. This skill guides scanning, detection, package setup, import, grouping, and validation in FiftyOne.

## Install

```bash
npx skillstore add adonaivera/fiftyone-dataset-import
```

## Metadata

- Status: approved
- Slug: adonaivera-fiftyone-dataset-import
- Skillstore revision: r1
- Version status: missing
- Tree hash: ae33cc14591fb1070c0b340a2249b9580eeeeeb9aa39ab58cd10f6774149ed9a
- Author: AdonaiVera
- GitHub username: AdonaiVera
- License: MIT
- Repository: https://github.com/AdonaiVera/fiftyone-skills/tree/main/dataset-import/skills/fiftyone-dataset-import
- Ref: 6eec4a59f63ff83440a84c18d1aef79743bc714e
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network, filesystem
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/adonaivera-fiftyone-dataset-import
- Manifest: https://skillstore.pages.dev/api/skills/adonaivera-fiftyone-dataset-import/manifest

## Capabilities

- Scans dataset folders and inventories media, labels, and directory structure.
- Maps common formats such as COCO, YOLO, VOC, CVAT, KITTI, and OpenLABEL to FiftyOne settings.
- Guides grouped dataset creation for multi-camera, LiDAR, and mixed media workflows.
- Identifies required devkit packages and asks before installation.
- Validates imported sample counts and opens the FiftyOne App.
- Provides Python examples for specialized autonomous driving datasets.

## Use Cases

- Import Standard Detection Datasets: Load image datasets with COCO, YOLO, VOC, CVAT, or KITTI labels into a new FiftyOne dataset.
- Create Multimodal Driving Datasets: Group camera images, LiDAR point clouds, and labels by scene for autonomous driving review.
- Validate Dataset Migration Results: Compare imported sample counts, labels, groups, and slices against the source files before analysis.

## Prompt Templates

### Import a Simple Image Dataset

```
Scan my image dataset at [path], detect labels, propose a FiftyOne import plan, and wait for my confirmation.
```

### Detect Unknown Dataset Format

```
Inspect [path], identify the media types and annotation format, then recommend the safest FiftyOne import settings.
```

### Create Multimodal Groups

```
Analyze [path] for camera, LiDAR, and scene grouping patterns. Propose group slices and a validation checklist.
```

### Plan a Custom Autonomous Driving Import

```
Build a FiftyOne import plan for [dataset name] at [path], including required packages, conversion steps, and label mapping.
```

## Limitations

- Requires a running FiftyOne MCP server and Voxel51 plugins.
- Specialized formats may require third-party devkit packages.
- Custom import examples must be verified against the real dataset schema.
- Pickle examples should only be used with trusted files or replaced with safe parsers.

## Best Practices

- Scan the folder and present findings before creating a dataset.
- Ask for confirmation before installing packages or overwriting datasets.
- Validate sample counts, labels, and groups after every import.

## Anti Patterns

- Do not import before confirming detected formats with the user.
- Do not flatten multimodal scenes that should stay grouped.
- Do not load untrusted pickle files without a safe review process.

## Security Audit

- Audited at: 2026-07-05T01:22:29.982\+00:00
- Summary: Most static hits are Markdown backticks, code fences, examples, or benign FiftyOne references. Confirmed issues are limited to package installation guidance and Python import checks that can execute third-party code. A high-risk semantic issue remains because the skill recommends pickle.load on dataset files.

## Stats

- Views: 378
- Downloads: 11
- Favorites: 0
- Popularity score: 0