LazyLabel

AI-Assisted Image Segmentation for Machine Learning Dataset Preparation

LazyLabel combines Meta's Segment Anything Model (SAM) with comprehensive manual annotation tools to accelerate the creation of pixel-perfect segmentation masks for computer vision applications.

---

Quick Start

pip install lazylabel-gui
lazylabel-gui

From source:

git clone https://github.com/dnzckn/LazyLabel.git
cd LazyLabel
pip install -e .
lazylabel-gui

Requirements: Python 3.10+, 8GB RAM, ~2.5GB disk space (for model weights)

---

Core Features

AI-Powered Segmentation

LazyLabel leverages Meta's SAM for intelligent object detection:

• Single-click object segmentation
• Interactive refinement with positive/negative points
• Support for both SAM 1.0 and SAM 2.1 models
• GPU acceleration with automatic CPU fallback

Manual Annotation Tools

When precision matters:

• Polygon drawing with vertex-level editing
• Bounding box annotations for object detection
• Edit mode for adjusting existing segments
• Merge tool for combining related segments

Image Processing & Filtering

Advanced preprocessing capabilities:

• FFT filtering: Remove noise and enhance edges
• Channel thresholding: Isolate objects by color
• Border cropping: Define crop regions that set pixels outside the area to zero in saved outputs
• View adjustments: Brightness, contrast, gamma correction

Multi-View Mode

Process multiple images efficiently:

• Annotate up to 4 images simultaneously
• Synchronized zoom and pan across views
• Mirror annotations to all linked images

---

Export Formats

NPZ Format (Semantic Segmentation)

One-hot encoded masks optimized for deep learning:

import numpy as np

data = np.load('image.npz')
mask = data['mask']  # Shape: (height, width, num_classes)

# Each channel represents one class
sky = mask[:, :, 0]
boats = mask[:, :, 1]
cats = mask[:, :, 2]
dogs = mask[:, :, 3]

YOLO Format (Object Detection)

Normalized polygon coordinates for YOLO training:

0 0.234 0.456 0.289 0.478 0.301 0.523 ...
1 0.567 0.123 0.598 0.145 0.612 0.189 ...

Class Aliases (JSON)

Maintains consistent class naming across datasets:

{
  "0": "background",
  "1": "person",
  "2": "vehicle"
}

---

Typical Workflow

1. Open folder containing your images
2. Click objects to generate AI masks (mode 1)
3. Refine with additional points or manual tools
4. Assign classes and organize in the class table
5. Export as NPZ or YOLO format

Advanced Preprocessing Workflow

For challenging images:

1. Apply FFT filtering to reduce noise
2. Use channel thresholding to isolate color ranges
3. Enable "Operate on View" to pass filtered images to SAM
4. Fine-tune with manual tools

---

Advanced Features

Multi-View Mode

Access via the "Multi" tab to process multiple images:

• 2-view (side-by-side) or 4-view (grid) layouts
• Annotations mirror across linked views automatically
• Synchronized zoom maintains alignment

SAM 2.1 Support

LazyLabel supports both SAM 1.0 (default) and SAM 2.1 models. SAM 2.1 offers improved segmentation accuracy and better handling of complex boundaries.

To use SAM 2.1 models:

1. Install the SAM 2 package:

   pip install git+https://github.com/facebookresearch/sam2.git

2. Download a SAM 2.1 model (e.g., sam2.1_hiera_large.pt) from the SAM 2 repository
3. Place the model file in LazyLabel's models folder:
   • If installed via pip: ~/.local/share/lazylabel/models/ (or equivalent on your system)
   • If running from source: src/lazylabel/models/
4. Select the SAM 2.1 model from the dropdown in LazyLabel's settings

Note: SAM 1.0 models are automatically downloaded on first use.

---

Key Shortcuts

Action	Key	Description
AI Mode	1	SAM point-click segmentation
Draw Mode	2	Manual polygon creation
Edit Mode	E	Modify existing segments
Accept AI Segment	Space	Confirm AI segment suggestion
Save	Enter	Save annotations
Merge	M	Combine selected segments
Pan Mode	Q	Enter pan mode
Pan	WASD	Navigate image
Delete	V/Delete	Remove segments
Undo/Redo	Ctrl+Z/Y	Action history

---

Building Windows Executable

Create a standalone Windows executable with bundled models (no Python required):

Requirements:

• Windows (native, not WSL)
• Python 3.10+
• PyInstaller: pip install pyinstaller

Build steps:

git clone https://github.com/dnzckn/LazyLabel.git
cd LazyLabel
python build_system/windows/build_windows.py

The executable will be created in dist/LazyLabel/. The entire folder (~7-8GB) can be moved anywhere and runs offline.

---

Documentation

• Usage Manual - Comprehensive feature guide
• Architecture Guide - Technical implementation details
• GitHub Issues - Report bugs or request features

---