VueFusion
True next-generation tracking technology.
- State of the Art ML Models
- Household Object Tracking: Chairs, Cups, etc.
- Intel, AMD and NVIDIA supported via DirectML
- Inferencing latency as low as 3.5ms
- Accurate Multi-Camera Depth Triangulation
- Open Source under AGPL v3.0
Find it on:
How can I help?
- Model Training Data: Set up a phone and record yourself playing VR. The more variety in poses, backgrounds & clothing, the better the models will end up. Training data will be anonymized to a reasonable extent and published to an open dataset. Do not send in videos that you do not want others to see.
- Development: Are you familiar with Kotlin, Python, OpenCV, or 3D projection? I'd love any contributions you are willing to make!
- Marketing: Got any fun or unique ideas for posters? Draw one up and send it in! Even just telling others this project exists helps.
- Financial: Making software like this takes a lot of time and love, help keep me going by sending a donation to my Ko-Fi.
Join the Discord for more information.
What makes this better than MediaPipe Pose or similar?
- Kalman Filtering for continuous prediction and no camera sync requirements
- Fisheye Lens model for >130° camera FOV
- One Euro Filtering for low-latency smoothing
- Coroutine based architecture for limitless scalability
- Modular architecture for implementing non-AI detectors like AprilTags
- Ability to swap out models and train your own
- Synthetic data pipeline built on Blender for easy model training
- HiDPI & Touch friendly user interface
- // TODO: Add more reasons here