wxcvannotator 0.1.0

Professional computer vision annotation tool with AI assistance

English | 繁體中文

wxCvAnnotator

Industrial-Grade AI-Assisted Image Annotation Tool

wxCvAnnotator is a high-performance desktop annotation tool built on wxPython and a C++ OpenCV engine (wxCvModule). It is optimized for industrial vision workflows, offering smooth navigation of high-resolution images alongside AI-powered annotation assistance.

---

✨ Features

Annotation Tools

A complete set of geometric drawing tools for various computer vision tasks:

Tool	Description	Use Case
Rectangle	Standard bounding box	Object Detection
Rotated Rectangle	Angle-aware bounding box	Industrial AOI, OBB
Polygon	Arbitrary shape contour	Instance Segmentation
Circle / Annulus	Circular and ring-shaped annotations	Geometric Feature Labeling
Point	Single keypoint	Pose Estimation

AI-Assisted Annotation

Integrated deep learning models to dramatically accelerate labeling:

• Segment Anything (SAM / SAM 2 / EfficientSAM): Click on an object to instantly generate a precise mask and polygon contour. ONNX quantized models (40 MB–600 MB) are automatically downloaded from HuggingFace on first use.

Data Formats & Management

• Storage: LabelMe JSON format by default (one .json per image), fully compatible with existing LabelMe datasets.
• Export: YOLO TXT, COCO JSON, Pascal VOC XML, CSV.
• Label Management: Three-level label loading (Project → Global → Default), custom colors, and Batch Rename.
• Attribute Panel: Attach flags (boolean), custom key-value attributes, and transcription text to each annotation.

Interface & Experience

• Large Image Support: C++ rendering engine optimized for high-resolution images — smooth zoom and pan.
• Three-Panel Layout: Toolbar (left) → Canvas (center) → Instance List (right).
• Resizable Panels: Drag splitters to adjust panel widths; layout is remembered across sessions.
• Multilingual: UI supports English, Traditional Chinese, and Japanese.
• Theme Switching: Built-in Light / Dark themes with custom color scheme support.

---

📦 Installation

Prerequisites

All platforms require wxCvModule (the C++ rendering engine) to be installed first:

# Install from PyPI (if a wheel is available for your platform)
pip install wxcvmodule

# Or install manually from the Releases page
pip install wxcvmodule-*.whl

If wxCvModule is not installed, the application will start in Mock mode — some C++ rendering features will be disabled, but the Python UI remains functional.

---

Windows

Tested on: Windows 10/11 x64

1. Install Python 3.10+

Download from python.org. Make sure to check Add Python to PATH during installation.

2. Install wxCvModule

pip install wxcvmodule

3. Install wxCvAnnotator

pip install wxcvannotator

4. Launch

wxcv-annotator

Optional — NVIDIA GPU acceleration:

pip install onnxruntime-gpu

---

macOS

Tested on: macOS arm64 (Apple Silicon), Python 3.12

1. Install pyenv (recommended)

brew install pyenv

Add to ~/.zshrc:

export PYENV_ROOT="$HOME/.pyenv"
export PATH="$PYENV_ROOT/bin:$PATH"
eval "$(pyenv init -)"

2. Install Python 3.12

pyenv install 3.12.12
pyenv local 3.12.12

3. Install wxCvModule

pip install wxcvmodule
# If no macOS wheel is available on PyPI, download manually from the Releases page:
pip install wxcvmodule-*.macosx-*.whl

4. Install and launch

pip install wxcvannotator
wxcv-annotator

macOS notes:

• objc[] duplicate class warnings at startup are a known issue and do not affect functionality — they can be safely ignored.
• Do not run python -c "import wxCvModule" directly in the shell; always launch via wxcv-annotator.

---

Linux

Tested on: Ubuntu 22.04 x64

1. Install system dependencies

sudo apt-get update
sudo apt-get install -y python3.12 python3.12-pip python3-dev \
    libgtk-3-dev libgl1-mesa-glx libglib2.0-0

2. Install wxCvModule

pip install wxcvmodule

3. Install and launch

pip install wxcvannotator
wxcv-annotator

Optional — NVIDIA CUDA GPU acceleration:

pip install onnxruntime-gpu

---

Developer Install

git clone https://github.com/wxCvRoot/wxCvAnnotator.git
cd wxCvAnnotator
pip install -e .
wxcv-annotator

---

Launch Options

wxcv-annotator                       # Standard launch (after install)
wxcv-annotator /path/to/image.jpg   # Open a specific image
wxcv-annotator /path/to/folder/     # Open a folder
python -m wxcvannotator             # Module mode
python main.py                      # Dev launcher

---

Verify Installation

python -c "
import wx; print(f'wxPython: {wx.version()}')
import numpy; print(f'numpy: {numpy.__version__}')
import cv2; print(f'opencv: {cv2.__version__}')
try:
    import onnxruntime; print(f'onnxruntime: {onnxruntime.__version__}')
except ImportError:
    print('onnxruntime: not installed (AI features disabled)')
import wx; import wxCvModule; print('wxCvModule: OK')
import wxcvannotator; print(f'wxcvannotator: {wxcvannotator.__version__}')
"

---

🔄 Basic Workflow

1. Open image: Go to File > Open Image or File > Open Folder, or drag and drop an image onto the window.
2. Select tool: Click an annotation tool in the left toolbar (Rectangle, Polygon, Rotated Rect, etc.).
3. Draw: Click or drag on the canvas to draw the shape.
4. Add to list: Press Enter or click [➕ Add] in the toolbar. Set the label in the dialog and confirm.
5. Edit: Click an annotation in the right-side list, adjust it on the canvas, then press Enter to update.
6. Attributes: With an annotation selected, use the bottom panel to edit flags, custom key-value attributes, and transcription text.
7. Auto-save: Annotations are automatically saved as a .json file (LabelMe format) alongside the image.

---

🤖 AI-Assisted Annotation

Workflow

1. Select AI Polygon or AI Mask from the left toolbar.
2. Choose a model from the Model dropdown:
   • EfficientSAM — Speed-first; ideal for real-time annotation.
   • SAM / SAM 2 — Accuracy-first; best for complex boundaries.
3. Click on the image to add prompt points:
   • Left-click (green dot): Positive prompt — include this area.
   • Right-click (red dot): Negative prompt — exclude this area.
   • Backspace: Remove the last prompt point.
4. A preview polygon updates in real time after each click.
5. Finalize:
   • Press Enter or double left-click: Confirm and add to the list.
   • Double right-click: Remove the last prompt point.

First use: The selected model will be automatically downloaded from HuggingFace when you first click. Ensure you have an internet connection. Download progress is shown in the status bar.

---

📤 Dataset Export

Single Image Export

Go to Export > Export Current Image..., choose the target format, and confirm. The output file is saved in the same directory as the image.

Dataset Split Export

Go to Export > Export Dataset (Split)... to configure:

• Honor manual status: Images marked as train/val/test in the file list are assigned to those splits first.
• Adjust ratios: Set train/val/test split ratios for unmarked images using sliders.
• Copy images: Choose whether to copy the actual image files to produce a standalone dataset package.

Supported formats:

Format	Description
YOLO TXT	YOLOv5/v8/v11 — supports det / seg / obb
COCO JSON	Standard COCO format — supports detection and segmentation
Pascal VOC XML	Classic XML format for Detectron2 and similar frameworks
CSV	Lightweight general-purpose format

---

⌨️ Keyboard Shortcuts

Key	Action
S	Switch to Select / Edit mode
P	Polygon tool
R	Rotated Rectangle tool
Enter	Add / Update annotation
Esc	Cancel current drawing
Delete	Delete selected annotation
Scroll wheel	Zoom image
Middle-click / Hold Space	Pan image
Double left-click (AI mode)	Confirm AI prediction
Double right-click (AI mode)	Remove last prompt point

---

🛠️ System Requirements

Item	Minimum
Python	3.10+
wxPython	4.2.0+
OS	Windows 10+, macOS 12+, Ubuntu 20.04+
RAM	4 GB (8 GB+ recommended for AI features)
Disk	< 100 MB base; AI models add 40 MB–600 MB

---

🔗 FAQ

Q: wxcv-annotator command not found after install?

Make sure Python's Scripts directory is in your PATH. On Windows, try python -m wxcvannotator.

Q: ModuleNotFoundError: No module named 'wx'?

pip install wxPython

Q: AI annotation is unavailable?

Install onnxruntime:

pip install onnxruntime      # CPU
pip install onnxruntime-gpu  # NVIDIA GPU

Q: Application starts in Mock mode?

wxCvModule is not correctly installed. Verify with:

pip show wxcvmodule

Q: objc[] duplicate class warnings on macOS?

This is a known issue caused by both cv2 and wxCvModule bundling their own OpenCV dynamic libraries. It does not affect functionality and can be ignored.

---

⚖️ License

This project is licensed under the Apache License 2.0.

Copyright 2026 wxCvRoot

See the LICENSE file for full terms.

This project makes use of the following open-source components:

• OpenCV — Apache 2.0
• wxWidgets — wxWindows Licence
• Segment Anything (SAM) — Apache 2.0
• EfficientSAM — Apache 2.0

---

Maintained by wxCvRoot