CycleISP环境配置

配置

环境：AutoDL GPU服务器

https://github.com/swz30/CycleISP

git clone https://github.com/swz30/CycleISP.git

sudo apt-get install cmake build-essential libjpeg-dev libpng-dev

pip install matplotlib scikit-image yacs lycon natsort h5py tqdm

pip install opencv-python opencv-contrib-python -i https://pypi.tuna.tsinghua.edu.cn/simple

问题：

*.Ubuntu ModuleNotFoundError: No module named 'skimage.measure.simple_metrics'

--->pip install scikit-image==0.16.2

python debug:cannot import name ‘compare_psnr‘& cannot import name ‘compare_ssim‘_呆呆象呆呆的博客-CSDN博客

*.Ubuntu python generate_rgb_data.py报错

--->

*.Win10 numpy报错：Original error was: DLL load failed: 找不到指定的模块。

--->pip install --upgrade numpy（或者卸载重新安装）

*.Win10  lycon安装报错

--->cv2替换lycon操作

*.Win10  加载数据使用多线程报错

 --->

 

代码+原理

*.torch实现高斯核

二维高斯核公式如下：

C语言实现便于理解，可以debug打印直观的看下高斯核

/*************************************************************
*Function: Gauss mask compute
*Params:
*      r-radius of gauss filter
*      sigma-sigma of gauss filter
*      gaussMask-gauss weight to compute with size of (2r+1)*(2r+1)
*Return NULL
************************************************************/
void GaussMask(int r, double sigma, double gaussMask[])
{
	double PI = 3.1415926;
	double sum = 0;
	int stride = 2 * r + 1;
    for (int y = -r, h = 0; y <= r; y++, h++)
    {
        for (int x = -r, w = 0; x <= r; x++, w++)
        {
            gaussMask[w + h * stride] = (1.0 / (2.0 * PI * sigma * sigma)) * (exp(-((double)x * (double)x + (double)y * (double)y) / (2.0 * sigma * sigma)));
			sum += gaussMask[w + h * stride];
        }
    }
	for (int i = 0; i < stride * stride; i++)
    {
        gaussMask[i] = gaussMask[i] / sum;
    }
};

*.噪声模型

*.模型结构