GitHub Repo

Embeddings Visualization#

Visualize your dataset in 2D using UMAP or t-SNE to explore structure, find clusters, and identify outliers.

Install#

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

When prompted, select fiftyone-embeddings-visualization from the menu.

Requirements#

Usage#

Load a dataset in FiftyOne, then ask your AI assistant:

"Visualize my dataset embeddings with UMAP"
"Compute CLIP embeddings and show the 2D projection"
"Color the embedding plot by class label and find outliers"

The skill computes embeddings if they don’t exist, reduces them to 2D, and opens the Embeddings panel in the FiftyOne App.

Example#

import fiftyone as fo
import fiftyone.zoo as foz

dataset = foz.load_zoo_dataset("quickstart")

Then ask your assistant:

"Compute CLIP embeddings for quickstart and show a UMAP visualization"

See also#