cuda-skimage的安装与简单入门

如何在windows上安装cucim?

  从rapids的官网上将项目下载并解压，然后把cucim-branch下的version文件复制到python下的cucim文件夹中并覆盖，

然后复制到src的cucim文件夹做同样操作即可。

  接在在python下的cucim文件夹里找到pyproject.toml文件，打开并查看安装所需要的环境。当环境符合的时候就可以cmd到python的cucim文件夹下执行安装命令python setup.py build和python setup.py install，至此cucim的安装应该是完成了。

cucim和skimage的简单对比

from cucim.skimage import measure,color
import cupy as cp
from time import time
from skimage import data,filters,io
from skimage.measure import label,regionprops
from matplotlib import pyplot as plt

img=io.imread("1.png")

cuimg=cp.asarray(img)

start=time()
labels=measure.label(cuimg,connectivity=2)
mid=time()
properties=measure.regionprops(labels)
end=time()

print(end-mid,mid-start)

start=time()
labels=label(img,connectivity=2)
mid=time()
properties=regionprops(labels)
end=time()
print(end-mid,mid-start)

'''
11.306145429611206 0.17850136756896973
32.30134105682373 1.1464252471923828
'''
cudst = color.label2rgb(labels, bg_label=0)
dst=cp.asnumpy(cudst)
plt.imshow(dst,interpolation='nearest')
plt.show()