Note: This project is in active development.

Disclaimer: This application is not intended for use in environments where it violates the respective Terms of Service. The developer is not responsible for any misuse.

About

SightKeeper is a desktop application for Windows and Linux that handles the complete computer vision pipeline for gamified environments, powered by state-of-the-art models.

How It Works

1. Capture: Utilizes low-level OS APIs (DirectX Desktop Duplication on Windows, MIT-SHM on Linux X11) for high-performance screen capture.
2. Annotate: Features a custom-built annotation tools for labeling captured images to create datasets for model training.
3. Train: Orchestrates the training of object detection models via local Python/Conda environments.
4. Deploy: Exports trained models to ONNX format for efficient inference using ONNX Runtime, supporting both CPU and GPU via CUDA.

Showcase