FiftyOne - Dataset Visualization & Curation

Overview

The

fiftyone

Key Concept: FiftyOne runs as a multi-container application with a MongoDB sidecar. The main FiftyOne container handles the web UI and processing, while MongoDB stores dataset metadata.

Quick Reference

ujust fiftyone config [--port=...] [--bind=...]

ujust fiftyone start [--instance=N|all]

ujust fiftyone stop [--instance=N|all]

ujust fiftyone restart [--instance=N|all]

ujust fiftyone logs [--instance=N] [--lines=...]

ujust fiftyone status [--instance=N]

ujust fiftyone url [--instance=N]

ujust fiftyone shell [--instance=N] [-- CMD...]

ujust fiftyone plugins [-- CMD...]

ujust fiftyone delete [--instance=N|all]

Parameters

--config-dir

-c

~/.config/fiftyone/{N}

--workspace-dir

-w

""

--bind

-b

127.0.0.1

--port

-p

5151

--image

-i

docker.io/voxel51/fiftyone

--tag

-t

latest

--gpu-type

-g

auto

--lines

-l

50

--instance

-n

1

Configuration

# Default configuration (port 5151, localhost only)
ujust fiftyone config

# Custom port (long form)
ujust fiftyone config --port=5152

# Custom port (short form)
ujust fiftyone config -p 5152

# Network-wide access
ujust fiftyone config --bind=0.0.0.0

# With workspace mount
ujust fiftyone config --workspace-dir=/data/datasets

# Combine parameters (long form)
ujust fiftyone config --port=5152 --bind=0.0.0.0 --workspace-dir=/data

# Combine parameters (short form)
ujust fiftyone config -p 5152 -b 0.0.0.0 -w /data

Update Existing Configuration

Running

config

Lifecycle Commands

Start/Stop/Restart

# Start FiftyOne (includes MongoDB sidecar)
ujust fiftyone start

# Start specific instance (long form)
ujust fiftyone start --instance=1

# Start specific instance (short form)
ujust fiftyone start -n 1

# Start all instances
ujust fiftyone start --instance=all

# Stop FiftyOne + MongoDB
ujust fiftyone stop --instance=1

# Restart all containers
ujust fiftyone restart

View Logs

FiftyOne shows interleaved logs from both the main container and MongoDB sidecar:

# Follow logs (default 50 lines)
ujust fiftyone logs

# More lines (long form)
ujust fiftyone logs --lines=100

# More lines (short form)
ujust fiftyone logs -l 100

# Specific instance
ujust fiftyone logs -n 1 -l 100

Log output format:

[fiftyone-mongodb] 2024-01-09 10:00:01 MongoDB started
[fiftyone] 2024-01-09 10:00:02 Connecting to database...
[fiftyone] 2024-01-09 10:00:03 FiftyOne App ready on port 5151

Get URL

ujust fiftyone url
# Output: http://localhost:5151

# Specific instance
ujust fiftyone url --instance=2

Shell Access

# Interactive shell
ujust fiftyone shell

# Run specific command (use -- separator)
ujust fiftyone shell -- fiftyone --version
ujust fiftyone shell -- pip list

# Specific instance
ujust fiftyone shell --instance=2 -- python -c "import fiftyone as fo; print(fo.__version__)"

# Short form
ujust fiftyone shell -n 2 -- ls -la

Plugin Management

# List installed plugins
ujust fiftyone plugins -- list

# Install a plugin
ujust fiftyone plugins -- install <plugin-name>

# Update plugins
ujust fiftyone plugins -- update

Multi-Container Architecture

FiftyOne runs with a MongoDB sidecar:

+-------------------+        +-------------------+
|    FiftyOne       |        |     MongoDB       |
|   (fiftyone-1)    | -----> | (fiftyone-mongodb-1) |
|   Port 5151       |        |   Port 27017      |
+-------------------+        +-------------------+
         |                            |
         +---- bazzite-ai network ----+

Container Names:

• fiftyone-{N}

  - Main FiftyOne container

• fiftyone-mongodb-{N}

  - MongoDB sidecar

Lifecycle:

• start

  starts MongoDB first, then FiftyOne

• stop

  stops FiftyOne first, then MongoDB

• logs

  shows interleaved output from both

Port Allocation

GPU Support

FiftyOne supports GPU acceleration for ML model inference:

# Auto-detect GPU (default)
ujust fiftyone config

# Explicit NVIDIA (long form)
ujust fiftyone config --gpu-type=nvidia

# Explicit NVIDIA (short form)
ujust fiftyone config -g nvidia

Configuration Files

~/.config/fiftyone/instance-{N}.env

~/.config/containers/systemd/fiftyone-{N}.container

~/.config/containers/systemd/fiftyone-mongodb-{N}.container

Common Workflows

Initial Setup

# 1. Configure FiftyOne with dataset directory
ujust fiftyone config --workspace-dir=/data/datasets

# 2. Start FiftyOne
ujust fiftyone start

# 3. Get URL
ujust fiftyone url

# 4. Open in browser
# http://localhost:5151

Dataset Analysis

# Start FiftyOne
ujust fiftyone start

# Open shell for interactive work
ujust fiftyone shell

# Inside container:
# import fiftyone as fo
# dataset = fo.load_dataset("my_dataset")
# session = fo.launch_app(dataset)

Network Access

# Configure for network access
ujust fiftyone config --bind=0.0.0.0

# Restart to apply
ujust fiftyone restart

# Access from other machines
# http://<hostname>:5151

Troubleshooting

FiftyOne Won't Start

Check:

ujust fiftyone status
ujust fiftyone logs --lines=50

Common causes:

• Port 5151 already in use
• MongoDB failed to start
• GPU driver issues

Fix:

# Delete and reconfigure
ujust fiftyone delete
ujust fiftyone config --port=5152
ujust fiftyone start

MongoDB Connection Failed

Symptom: FiftyOne logs show "Connection refused" to MongoDB

Check:

# Check MongoDB container
podman ps | grep fiftyone-mongodb
ujust fiftyone logs | grep mongodb

Fix:

# Restart both containers
ujust fiftyone restart

Datasets Not Persisting

Symptom: Datasets disappear after restart

Check:

• Verify config_dir is properly set
• Check MongoDB volume mounts

Fix:

# Reconfigure with explicit config directory
ujust fiftyone config --config-dir=/data/fiftyone

Cross-References

• Related Skills:

  jupyter

  (ML notebooks),

  ollama

  (LLM inference)
• FiftyOne Docs: https://docs.voxel51.com/
• GPU Setup:

  ujust config gpu setup

When to Use This Skill

Use when the user asks about:

• "fiftyone", "dataset visualization", "dataset curation"
• "ML datasets", "computer vision datasets"
• "data labeling", "annotation tool"
• "start fiftyone", "configure fiftyone"