swiftlabel 0.1.0

Keyboard-first image classification tool for ML practitioners

SwiftLabel

Keyboard - first image classification tool for ML practitioners who value speed and simplicity.

SwiftLabel is a local-only, zero-config tool that lets you classify images using keyboard shortcuts. All changes are staged until you commit them, ensuring a non-destructive workflow.

Features

• Keyboard-driven workflow - Hands never leave the keyboard
• Local-first & private - Images never leave your machine
• Non-destructive - Preview all changes before committing
• Zero config - Open folder, define classes, start labeling
• Session persistence - Resume where you left off
• Undo support - Revert your last actions

Installation

pip install git+https://github.com/ksavkin/SwiftLabel.git

Or install from source:

git clone https://github.com/ksavkin/SwiftLabel.git
cd swiftlabel
pip install -e ".[dev]"

Quick Start

# Basic usage (try with included examples!)
swiftlabel ./examples --classes cat,dog,bird

# With custom port
swiftlabel ./images --classes cat,dog,bird --port 9000

# Without auto-opening browser
swiftlabel ./images --classes cat,dog,bird --no-browser

A browser window will open at http://localhost:8765 where you can start labeling.

Remote Usage (SSH)

To use SwiftLabel on a remote server (e.g., a training machine with your dataset), you can forward the port via SSH.

1. On your local machine, create an SSH tunnel:

   # Forward local port 8001 to remote port 8001
   ssh -L 8001:localhost:8001 user@remote-host
   

2. On the remote server, run SwiftLabel on the forwarded port:

   # Run with --no-browser since you're on a remote terminal
   # Use python3 -m if the 'swiftlabel' command is not in your PATH
   python3 -m swiftlabel.cli ./your_dataset --classes up,down --port 8001 --no-browser
   

3. On your local machine, open http://localhost:8001 in your browser.

Restarting & Stopping

If you see [Errno 98] address already in use, it means a previous instance is still running.

• To stop a running server: Press Ctrl+C in the terminal.
• To force-kill a stale server (if Ctrl+C doesn't work or port is stuck):

  # Replace 8765 with your port number
  lsof -ti:8765 | xargs kill -9
  

Troubleshooting Remote Usage

• Port already in use: If you see [Errno 98] address already in use, find and kill the process:

  lsof -ti:8001 | xargs kill -9
  

• WebSocket connection fails: If you see a TypeError regarding logger in websockets, upgrade the library:

  pip install --upgrade websockets
  

Keyboard Shortcuts

Basic Controls

Key	Action
1-9	Assign to class 1-9
0	Assign to class 10 (if available)
D	Mark for deletion
U	Undo last action
← / →	Previous / Next image
Enter	Commit all changes
Esc	Cancel / Close modal
?	Show help overlay

Vim-style Navigation

Key	Action
H / L	Previous / Next image
J / K	Next / Previous image

How It Works

1. Launch: Run swiftlabel ./images --classes cat,dog,bird
2. Label: Press 1, 2, or 3 to classify images
3. Review: Press Enter to preview pending changes
4. Commit: Confirm to apply changes to filesystem

Staging

All changes are staged until you commit:

• Labeled images will be moved to class folders (e.g., cat/image.jpg)
• Deleted images will be removed from disk
• You can undo any action before committing

Session Persistence

SwiftLabel automatically saves your progress to .swiftlabel/session.json. When you restart, it resumes from where you left off.

Data Organization

SwiftLabel supports two main workflows:

1. Labeling from Scratch (Flat Folder)

If you have a folder full of unsorted images:

my_dataset/
  ├── image1.jpg
  ├── image2.jpg
  └── ...

When you commit, SwiftLabel will create class folders and move images into them:

my_dataset/
  ├── cat/
  │   └── image1.jpg
  ├── dog/
  │   └── image2.jpg
  └── ...

2. Reviewing Existing Labels (Structured Folders)

If your images are already organized into folders:

my_dataset/
  ├── cat/
  │   └── image1.jpg  <-- Auto-detected as "cat"
  ├── dog/
  │   └── image2.jpg  <-- Auto-detected as "dog"

SwiftLabel will read the folder names as initial labels. You can freely change labels or delete images, and changes will be applied to the file system upon commit.

CLI Options

Usage: swiftlabel [OPTIONS] DIRECTORY

Arguments:
  DIRECTORY  Path to the folder containing images to classify

Options:
  -c, --classes TEXT   Comma-separated list of class names (required)
  -h, --host TEXT      Server host address [default: 127.0.0.1]
  -p, --port INTEGER   Server port [default: 8765]
  --no-browser         Don't open browser automatically
  --debug              Enable debug logging
  --version            Show version and exit
  --help               Show this message and exit

Supported Image Formats

• JPEG (.jpg, .jpeg)
• PNG (.png)
• WebP (.webp)
• GIF (.gif)
• BMP (.bmp)
• TIFF (.tiff, .tif)

Development

Setup

# Clone repository
git clone https://github.com/swiftlabel/swiftlabel
cd swiftlabel

# Install with dev dependencies
pip install -e ".[dev]"

Run Tests

# Run all tests
pytest tests/ -v

# Run with coverage
pytest tests/ -v --cov=swiftlabel --cov-report=term-missing

Type Checking

mypy swiftlabel/ --strict

Linting

ruff check swiftlabel/

API Reference

SwiftLabel exposes a REST API at http://localhost:8765/api:

Endpoint	Method	Description
/api/session	GET	Get full session state
/api/stats	GET	Get labeling statistics
/api/images	GET	List all images
/api/images/{id}	GET	Serve image file
/api/label	POST	Assign label to image
/api/delete	POST	Mark image for deletion
/api/undo	POST	Undo last action
/api/changes/preview	GET	Preview pending changes
/api/changes/commit	POST	Apply all changes

WebSocket endpoint at /ws for real-time updates.

Environment Variables

Variable	Default	Description
SWIFTLABEL_HOST	127.0.0.1	Server bind address
SWIFTLABEL_PORT	8765	Server port
SWIFTLABEL_DEBUG	false	Enable debug logging

License

MIT License. See LICENSE for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.