⚡ CUDA Planesweep Project

This project is part of the INFO-H503 course (ULB). It implements an optimized planesweep kernels in CUDA and benchmarks their performance on your GPU.

Please refer to doc/guidelines.pdf or doc/reportH503.pdf for more info.

Figure 1: Planesweep with Graph-cut

📜 Features

• CUDA kernels for Planesweep algorithm and argmin depth extraction.
• Prints GPU device info using CUDA runtime API.
• Benchmarks and pretty colours 🎨✨

⚙️ Installation

1. Clone the repository:

   git clone git@github.com:Ant0in/CUDA-Planesweep.git
   cd CUDA-Planesweep

2. Make sure you have Cuda installed and a Cuda-capable GPU. Cuda can be installed using pacman or apt:

   sudo pacman -Syu nvidia nvidia-utils nvidia-settings
   nvidia-smi  # check if you have a cuda-capable gpu
   sudo pacman -S cuda

You also need to make sure you have OpenCV installed. You can use pacman or apt as well.

3. Build using the provided Makefile:

   make  # you can use the clean rule to remove build files

Alternatively, you can compile it by hand using nvcc and cmake if you are on Windows.

🛠️ Usage

To run the planesweep algorithm, use:

./Planesweep min  # or use `gc` if you want to use grap-cut refinement

Make sure that Planesweep is in the same directory as the /res directory. Example of valid /res directory can be found at /res. It includes cam_params.json and cameras v_.png from 0 up to 15 at max.

Figure 2: Execution with argmin depth-refinement

📄 License

This project is licensed under the MIT License. See the LICENSE file for more details.

🙏 Acknowledgements

This project was developed for the GPU Computing course INFO—H503. Special thanks to Bonatto Daniele (ULB) and Soetens Eline (ULB) for their guidance and support.