GitHub Repo

Dataset Import#

Import any dataset into FiftyOne with automatic format detection. Supports local files, Hugging Face Hub, cloud storage, multimodal grouped data, and MCAP robotics/AV sensor recordings.

Install#

curl -sL skil.sh | sh -s -- voxel51/fiftyone-skills

When prompted, select fiftyone-dataset-import from the menu.

Requirements#

Usage#

Start the MCP server and ask your AI assistant:

"Import the COCO dataset from /path/to/data"
"Load the keremberke/license-plate-object-detection dataset from Hugging Face"
"Import this folder of images, there are cameras and LiDAR files grouped by scene"
"Import these MCAP recordings from /path/to/rosbags"

The skill scans your data, auto-detects the format and media types, and loads the dataset into FiftyOne. It handles images, videos, point clouds, MCAP multimodal recordings (robotics/AV sensor logs), COCO, YOLO, VOC, KITTI, and more without you specifying the format.

Example#

# Download sample COCO data to ~/fiftyone/coco-2017/validation
fiftyone zoo datasets download coco-2017 --split validation

Then ask your assistant:

"Import the COCO 2017 validation dataset from ~/fiftyone/coco-2017/validation and name it coco-2017-validation"

After the skill runs, verify in Python:

import fiftyone as fo
dataset = fo.load_dataset("coco-2017-validation")
print(dataset)

Or ask your assistant to open it in the App:

"Launch the FiftyOne App with the coco-2017-validation dataset"

See also#

This skill’s directory also ships reference files the skill reads on demand rather than every invocation: SPECIALIZED-3D-FORMATS.md (PandaSet, nuScenes, Waymo, Argoverse, KITTI 3D, Lyft L5, A2D2), USE-CASE-EXAMPLES.md (copy-paste starting points), and for MCAP recordings specifically, MCAP-TROUBLESHOOTING.md (why a tile isn’t rendering), MCAP-AUTHORING.md (authoring MCAP from raw sensor data, converting ROS bags, merging/patching), and MCAP-DATASET-AND-VALIDATION.md (capability flags and validating an import before calling it done).