Introduction to Makesense.ai
Makesense.ai is a versatile online platform designed for photo labeling, particularly beneficial for computer vision projects in deep learning. The platform is free to use and is accessible directly from a web browser, eliminating the need for any intricate installations. It functions seamlessly across different operating systems, making it a truly cross-platform solution. This tool is perfect for smaller projects, as it simplifies and speeds up the process of preparing datasets. Users can download labels in various supported formats. Developed using TypeScript and anchored by the React/Redux framework, makesense.ai offers a robust and user-friendly experience.
Advanced AI Integrations
In an effort to minimize time spent on photo labeling, makesense.ai integrates cutting-edge AI models to automate repetitive tasks and offer recommendations:
- YOLOv5: This is a powerful integration using yolov5js, enabling you to load pretrained models as well as your custom models trained with YOLOv5, converted to TensorFlow.js format.
- SSD: Pretrained on the COCO dataset, SSD aids in drawing bounding boxes and suggesting labels.
- PoseNet: This model estimates key body joints, allowing you to determine a person’s pose in images or video.
These AI functionalities are powered by TensorFlow.js, allowing the tool to operate swiftly while safeguarding user data privacy as images are processed directly on the device without needing to be uploaded.
Setup and Operation
Local Setup
For those preferring a local setup, the following steps are recommended:
- Clone the repository:
git clone https://github.com/SkalskiP/make-sense.git - Navigate to the main directory:
cd make-sense - Install dependencies with:
npm install - Start the app with hot reload:
npm start
It is important to use npm versions 8.x.x and Node.js versions v16.x.x to avoid functionality issues.
Docker Setup
Alternatively, you can set up makesense.ai using Docker:
- Build the Docker image:
docker build -t make-sense -f docker/Dockerfile . - Run the Docker image as a service:
docker run -dit -p 3000:3000 --restart=always --name=make-sense make-sense - Access the application at:
http://localhost:3000/
Keyboard Shortcuts and Functionality
The tool offers a range of keyboard shortcuts for enhanced productivity, such as polygon autocomplete, cancel polygon drawing, and navigation between images.
Import and Export Formats
Makesense.ai supports a variety of import and export formats for different labeling needs, including CSV, YOLO, VOC XML, VGG JSON, COCO JSON, and Pixel Mask. The application provides a thorough guide on how these formats are supported to ease the labeling process.
Privacy and Tutorials
User privacy is a priority for makesense.ai, as it processes images locally on the user’s device, ensuring that no images are sent to a server. For beginners, the platform offers tutorials to guide them in creating and training object detection models, making the pathway to mastering deep learning projects accessible.
Contribution and Licensing
The makesense.ai project is open to contributions and is licensed under the GPL-3.0 License. For users participating in related research, there is also a recommended citation format.
With this comprehensive tool, users are empowered to efficiently prepare datasets for deep learning applications, all within a secure and flexible online platform.