MiePy_CUDA

MiePy_CUDA is a Python module to calculate Mie scattering coefficients (backscatter and extinction coefficient) of spheres by CUDA. The implementation is extremely fast, about 0.045ms/case in large-scale scenarios. The implementation is modified by "Absorption and Scattering of Light by Small Particles (Bohren & Huffman)".

Dependencies

• CUDA >= 10.2
• PyTorch >= 1.0
• Python >= 3.6

Installation

• Install suitable CUDA Toolkit for compiling .cu files, and make sure CUDA_HOME in your system environment variable.
• Install PyTorch with CUDA support, because the implementation takes PyTorch as frontend.
• Install setuptools package in Python enviroment for execute the setup.py file.
• Execute python setup.py install to compile and install the package.
• Run python demo.py to verify the installation. The correct output is

tensor([0.0267, 0.0708, 0.2003], dtype=torch.float64) tensor([2.2240, 2.1679, 2.1110], dtype=torch.float64)
tensor([0.0042, 0.0042, 0.0042], dtype=torch.float64) tensor([2.0669, 2.0873, 2.1351], dtype=torch.float64)
tensor([0.0042, 0.0042, 0.0042], dtype=torch.float64) tensor([2.0215, 2.0281, 2.0445], dtype=torch.float64)

安装

• 安装合适的CUDA Toolkit用于编译.cu文件，并且确保CUDA_HOME存在于你的系统环境变量。
• 安装CUDA支持的PyTorch，因为我们采用它作为数据设备间交互的前端。
• 安装setuptools包到Python环境，它用于执行setup.py文件。
• 执行python setup.py install来编译和安装。
• 执行python demo.py来验证安装是否成功，正确的输出如下面所示：

tensor([0.0267, 0.0708, 0.2003], dtype=torch.float64) tensor([2.2240, 2.1679, 2.1110], dtype=torch.float64)
tensor([0.0042, 0.0042, 0.0042], dtype=torch.float64) tensor([2.0669, 2.0873, 2.1351], dtype=torch.float64)
tensor([0.0042, 0.0042, 0.0042], dtype=torch.float64) tensor([2.0215, 2.0281, 2.0445], dtype=torch.float64)