[英]Matlab vs Python 2D convolution performance
I like to prototype algorithms in Matlab, but I have the requirement of putting them on a server that also runs quite a bit of Python code. 我喜欢在Matlab中对算法进行原型设计,但我要求将它们放在也运行大量Python代码的服务器上。 Hence I quickly converted the code to Python and compared the two.
因此,我迅速将代码转换为Python并进行了比较。 The Matlab implementation runs ~1000 times faster (from timing function calls - no profiling).
Matlab实现的运行速度快了约1000倍(通过计时函数调用-无性能分析)。 Anyone know off hand why the performance of Python is so slow?
有人知道为什么Python的性能这么慢吗?
% init random data
w = 800;
h = 1200;
hmap = zeros(w,h);
npts = 250;
for i=1:npts
hmap(randi(w),randi(h)) = hmap(randi(w),randi(h))+1;
end
% Params
disksize = 251;
nBreaks = 25;
saturation = .9;
floorthresh =.05;
fh = fspecial('gaussian', disksize, disksize/7);
hmap = conv2(hmap, fh, 'same');
% Scaling, paritioning etc
hmap = hmap/(max(max(hmap)));
hmap(hmap<floorthresh) = 0;
hmap = round(nBreaks * hmap)/nBreaks;
hmap = hmap * (1/saturation);
% Show the image
imshow(hmap, [0,1])
colormap('jet')
import numpy as np
from scipy.signal import convolve2d as conv2
# Test data parameters
w = 800
h = 1200
npts = 250
# generate data
xvals = np.random.randint(w, size=npts)
yvals = np.random.randint(h, size=npts)
# Heatmap parameters
gaussianSize = 250
nbreaks = 25
# Preliminary function definitions
def populateMat(w, h, xvals, yvals):
container = np.zeros((w,h))
for idx in range(0,xvals.size):
x = xvals[idx]
y = yvals[idx]
container[x,y] += 1
return container
def makeGaussian(size, fwhm):
x = np.arange(0, size, 1, float)
y = x[:,np.newaxis]
x0 = y0 = size // 2
return np.exp(-4*np.log(2) * ((x-x0)**2 + (y-y0)**2) / fwhm**2)
# Create the data matrix
dmat = populateMat(w,h,xvals,yvals)
h = makeGaussian(gaussianSize, fwhm=gaussianSize/2)
# Convolve
dmat2 = conv2(dmat, h, mode='same')
# Scaling etc
dmat2 = dmat2 / dmat2.max()
dmat2 = np.round(nbreaks*dmat2)/nbreaks
# Show
imshow(dmat2)
Ok, problem solved for me thanks to suggestion from @Yves Daust's comments; 好了,感谢@Yves Daust的评论为我解决了问题;
The filter scipy.ndimage.filters.gaussian_filter
utilises the separability of the kernel and reduces the running time to within a single order of magnitude of the matlab implementation. 过滤器
scipy.ndimage.filters.gaussian_filter
利用内核的可分离性,并将运行时间减少到matlab实现的单个数量级内。
import numpy as np
from scipy.ndimage.filters import gaussian_filter as gaussian
# Test data parameters
w = 800
h = 1200
npts = 250
# generate data
xvals = np.random.randint(w, size=npts)
yvals = np.random.randint(h, size=npts)
# Heatmap parameters
gaussianSize = 250
nbreaks = 25
# Preliminary function definitions
def populateMat(w, h, xvals, yvals):
container = np.zeros((w,h))
for idx in range(0,xvals.size):
x = xvals[idx]
y = yvals[idx]
container[x,y] += 1
return container
# Create the data matrix
dmat = populateMat(w,h,xvals,yvals)
# Convolve
dmat2 = gaussian(dmat, gaussianSize/7)
# Scaling etc
dmat2 = dmat2 / dmat2.max()
dmat2 = np.round(nbreaks*dmat2)/nbreaks
# Show
imshow(dmat2)
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.