SAM2 Annotation Tool

中文 | English

An intelligent image segmentation annotation tool powered by Meta SAM2 (Segment Anything Model 2). Click to segment, zero-shot annotation of any object.

Features

Smart Segmentation

• Click Segmentation -- Left-click on a target area, SAM2 automatically generates segmentation boundaries
• Box Segmentation -- Drag a rectangle to trigger segmentation
• Polygon Drawing -- Click continuously to manually annotate complex boundaries, right-click to close

Annotation Management

• Real-time annotation list with select, delete, and visibility toggle
• Add custom text descriptions to each annotation
• 12 preset categories with hierarchical classification (animals/vehicles/buildings, etc.)
• Drag vertices to fine-tune boundaries, right-click menu to insert/delete vertices

Quality Assessment

• Built-in quality scoring system: comprehensive score, boundary smoothness, vertex density
• SAM2 confidence automatically recorded (average >94%)
• Configurable vertex density (low/medium/high)

Data Export

• COCO JSON -- Universal object detection/segmentation training
• Pascal VOC XML -- Traditional CV task compatibility
• Exports include quality metrics and confidence extension fields

Quick Start

Requirements

Component	Minimum	Recommended
Python	3.10+	3.12+
Memory	8 GB	16 GB+
GPU	None (CPU works)	NVIDIA 8 GB+

Installation

# Clone the repository
git clone https://github.com/YOUR_USERNAME/SAM2-Annotation-Tool.git
cd SAM2-Annotation-Tool

# Create a virtual environment
python -m venv venv
venv\Scripts\activate        # Windows
# source venv/bin/activate   # Linux/Mac

# Install dependencies
pip install -r requirements.txt

Configure the Model

Edit config.yaml to set the SAM2 model path:

model:
  path: "models/sam2_hiera_tiny.pt"
  config_path: "models/sam2_hiera_t.yaml"
  device: "cuda"  # use "cpu" if no GPU

Download model weights from the SAM2 official repository.

Launch

python main.py

Usage Workflow

Open image folder -> Select tool -> Click/box target -> SAM2 auto-segment
                                                          |
Export COCO/VOC <- Confirm annotation (add text) <- Fine-tune boundary (drag vertices)

Keyboard Shortcuts

• Ctrl+O Open folder
• Ctrl+S Save project
• Delete Delete selected annotation
• Mouse wheel Zoom canvas

Technical Architecture

+---------------------------------------------+
|         UI Layer (PyQt6)                    |
|   ImageList | CanvasWidget | LayerList      |
+---------------------------------------------+
|         Business Logic Layer                |
|   MainWindow | ImageManager | ProjectManager|
+---------------------------------------------+
|         Core Engine Layer                   |
|            SAM2Engine                       |
|   (PyTorch + SAM2 + OpenCV)                 |
+---------------------------------------------+
|         Data Layer                          |
|   Annotation | COCOExporter | VOCExporter   |
+---------------------------------------------+

Project Structure

SAM2-Annotation-Tool/
+-- main.py                 # Application entry point
+-- config.yaml             # Configuration
+-- requirements.txt        # Dependencies
+-- src/
|   +-- core/
|   |   +-- sam2_engine.py     # SAM2 segmentation engine
|   |   +-- annotation_data.py # Data models
|   |   +-- image_manager.py   # Image management
|   |   +-- project_manager.py # Project management
|   +-- gui/
|   |   +-- main_window.py     # Main window
|   |   +-- canvas_widget.py   # Canvas widget
|   |   +-- layer_list.py      # Annotation panel
|   |   +-- image_list.py      # Image list
|   |   +-- tool_bar.py        # Toolbar
|   +-- exporters/
|   |   +-- coco_exporter.py   # COCO export
|   |   +-- voc_exporter.py    # VOC export
|   +-- utils/
|       +-- config_loader.py   # Config loader
|       +-- file_utils.py      # File utilities
|       +-- image_utils.py     # Image utilities
+-- tests/
+-- docs/

Quality Metrics

The tool computes a quality score for each annotation:

quality_score = 0.5 * confidence + 0.3 * density_score + 0.2 * smoothness

Metric	Range	Description
quality_score	0--1	Comprehensive quality score
boundary_smoothness	0--1	Boundary smoothness
vertex_count	int	Number of polygon vertices
vertex_density	0--20	Vertex density (per 100px perimeter)
confidence	0--1	SAM2 model confidence

Comparison

Feature	SAM2 Annotation Tool	LabelImg	LabelMe	CVAT
Smart segmentation	SAM2-driven	Manual	Manual	Requires config
Zero-shot	Yes	No	No	No
Quality metrics	Built-in	None	None	Basic
Vertex density	Configurable	No	No	No
Local execution	Yes	Yes	Yes	Requires deploy
Learning curve	Low	Low	Medium	High

Dependencies

• PyQt6 >= 6.0.0
• PyTorch >= 2.0.0
• SAM2 >= 0.4.0
• OpenCV >= 4.8.0
• Pillow >= 10.0.0
• NumPy >= 1.24.0
• PyYAML >= 6.0

Documentation

• Product Introduction (Chinese)
• Technical Architecture (Chinese)

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for guidelines.

License

MIT License -- see LICENSE for details.

Acknowledgments

• Meta SAM2 -- The segmentation model powering this tool
• PyQt6 -- GUI framework
• OpenCV -- Image processing