fiftyone-embeddings-visualization

What problem does it solve?

Visualize high-dimensional dataset embeddings in 2D or 3D to reveal structure and clusters.

Core Features & Use Cases

• Embedding Visualization: Reduce embeddings to 2D/3D using UMAP, t-SNE, or PCA for interactive exploration.
• Exploratory Data Analysis: Compare classes, inspect cluster separation, and identify outliers in embedding space.
• Use Case: For example, visualize CIFAR-like dataset embeddings to observe class separation and detect mislabeled samples.

Quick Start

Use this Skill to visualize embeddings for your dataset:

• Set context: set_context(dataset_name="my-dataset")
• Launch the FiftyOne App: launch_app() and wait 5-10 seconds
• Ensure brain operators are available:
  • list_operators(builtin_only=False)
  • get_operator_schema(operator_uri="@voxel51/brain/compute_visualization")
• Compute embeddings if not present:
  • execute_operator( operator_uri="@voxel51/brain/compute_similarity", params={ "brain_key": "my_viz", "model": "clip-vit-base32-torch", "embeddings": "clip_embeddings", "backend": "sklearn", "metric": "cosine" } )
• Generate 2D visualization:
  • execute_operator( operator_uri="@voxel51/brain/compute_visualization", params={ "brain_key": "my_viz", "embeddings": "clip_embeddings", "method": "umap", "num_dims": 2 } )
• Open the App at http://localhost:5151/ and explore the Embeddings panel with brain key "my_viz".