Software installation#
We suggest to use Ubuntu 22.04 for the software stack. Since VisionPilot requires AI packages, we need to install the following packages first.
Prerequisites#
- CUDA: Optional for GPU processing.
- OpenCV: For image and video processing.
- Ubuntu:
sudo apt install libopencv-dev
- Ubuntu:
- ONNX Runtime: For model inference.
- Download from the GitHub release
- LibTorch: For tensor manipulation capabilities.
- Download from the PyTorch website
Models#
The following models will be used in the pipeline.
- SceneSeg - segmentation of foreground objects:
- Scene3D - monocular depth estimation
- DomainSeg - segmentation of construction objects