If you've ever built a computer vision model, you know this truth:
Data annotation is the slowest, most painful part of the pipeline.
You have thousands of images.
You need high-quality segmentation masks.
And your options usually look like this:
- Use a clunky, outdated desktop tool.
- Upload sensitive data to a cloud service and pay monthly.
- Spend hours manually outlining objects.
- Or build your own annotation tool from scratch.
There had to be a better way.
So we built one.
Meet NotumAi.
💡 Why We Built NotumAi
Modern segmentation models like Segment Anything Model 2 (SAM 2) from Meta can generate high-quality masks instantly.
But here’s the problem:
There isn’t a clean, developer-friendly, fully local tool that integrates these models into a smooth dataset creation workflow.
Most solutions are:
- Cloud-based
- Expensive
- Closed source
- Not optimized for custom pipelines
We wanted a tool that is:
🔒 100% Local — Your data never leaves your machine
⚡ GPU Accelerated — Real-time AI-assisted segmentation
🎨 Modern & Clean — Built for long annotation sessions
🌍 Open Source — Built with and for the community
That’s how NotumAi was born.
🛠️ What is NotumAi?
NotumAi is a professional-grade image annotation tool specifically designed for creating computer vision datasets. It combines a robust Python backend (handling the heavy lifting with PyTorch and SAM 2) with a modern Electron frontend (ensuring a responsive, beautiful interface).
Key Features
⚡ AI-Assisted Segmentation: Just click on an object, and NotumAi instantly generates a precise polygon mask using SAM 2.
🎨 Professional UI: A glassmorphism-inspired design with a focus on usability and aesthetics.
📂 Project Management: Organize your datasets, persist your progress, and manage multiple classes effortlessly.
💾 Flexible Export: Export your work in standard formats like COCO, YOLO, and Pascal VOC, ready for training.
🔒 Local & Secure: Your data stays with you. Perfect for sensitive or proprietary datasets.
🏗️ Under the Hood
For the developers out there, here's how we built it:
- Frontend: Electron, HTML5, Vanilla JS, and CSS (no heavy frameworks, just pure performance).
- Backend: Python with FastAPI and Uvicorn.
- AI Engine: PyTorch running Meta's SAM 2.1 model.
- Communication: Seamless HTTP communication between the frontend client and the local inference server
🌟 We Need You!
NotumAi is an Open Source project, and we are just getting started. We have a solid foundation, but to make this the ultimate annotation tool, we need the community's help.
Whether you are a:
- Frontend Wizard: Help us polish the UI, add new interaction modes, or optimize the canvas rendering.
- AI Engineer: Help us optimize the inference pipeline or integrate new models.
- Pythonista: detailed backend logic, better file handling, or new export formats. ...your contributions are welcome!
🔗 Get Involved
Check out the repository, give us a star, and let's build the best open-source annotation tool together.
Happy Annotating! 🖊️✨
Top comments (0)