我需要压缩大量(2000)的图像(2560x1440x1)在1位运行长度编码格式。(格式不是我的选择。)允许的最大运行长度是125次重复。你知道吗
我使用Python,并尝试使用numpython和本地Python。使用numpy的速度更快。在我切换到Cython之后,我的速度提高了200%,但是本地代码的速度比numpy快了(2倍)。你知道吗
赛顿200%的提速似乎太小了。我做错什么了吗?你知道吗
本机Python/Cython的主程序调用:
rlestack.append(rleEncode.encodedBitmap_Bytes_nonumpy(imgarr8.flatten(0).tolist()))
或者使用Numpy/Cython:
rlestack.append(rleEncode.encodedBitmap_Bytes_withnumpy(imgarr8.flatten(0)))
(imgarr8是一个2560x1440 numpy的数字.uint8)你知道吗
Cython RLE编码方法在RLE编码.pyx地址:
import numpy
cimport numpy
#!python
@cython.wraparound (False) #turn off negative indexing
@cython.boundscheck(False) # turn off bounds-checking
@cython.nonecheck(False)
def encodedBitmap_Bytes_nonumpy1D(list surfOrFile not None):
""" Converts image data from file on disk to RLE encoded byte string.
Encoding scheme:
Highest bit of each byte is color (black or white)
Lowest 7 bits of each byte is repetition of that color, with max of 125 / 0x7D
"""
#(width, height) = (1440, 2560)
cdef unsigned int nrPixels = 3686400
cdef unsigned int lastPixel = nrPixels - 1
# Count number of pixels with same color up until 0x7D/125 repetitions
rleData = bytearray() # convert bytearray to cdef array has no speed benefit
cdef unsigned char color = 0
cdef unsigned char prevColor = 0
cdef unsigned char black = 0
cdef unsigned char white = 1
cdef unsigned char nocolor=3
cdef unsigned char r
cdef unsigned char nrOfColor = 0
cdef unsigned char encValue = 0
cdef unsigned int pixelNr
cdef unsigned int isLastPixel = False
prevColor = nocolor
for pixelNr in range(nrPixels):
r = surfOrFile[pixelNr]
if (r and 0b10000000): #if (r<128)
color = white
else:
color = black
if prevColor == nocolor: prevColor = color
isLastPixel = (pixelNr == lastPixel)
if color == prevColor and nrOfColor < 0x7D and not isLastPixel:
nrOfColor = nrOfColor + 1
else:
# print (color,nrOfColor,nrOfColor<<1)
encValue = (prevColor << 7) | nrOfColor # push color (B/W) to highest bit and repetitions to lowest 7 bits.
rleData.append(encValue)
prevColor = color
nrOfColor = 1
return bytes(rleData)
#!python
@cython.boundscheck(False) # turn off bounds-checking
@cython.wraparound(False) # turn off bounds-checking
def encodedBitmap_Bytes_withnumpy(numpy.ndarray[numpy.npy_uint8,ndim=1] x):
# Encoding magic
cdef unsigned int n=0
cdef numpy.ndarray[numpy.npy_int64, ndim = 1] starts # npy_int64
cdef numpy.ndarray[numpy.npy_int64, ndim = 1] lengths# npy_int64
cdef numpy.ndarray[numpy.npy_uint8, ndim = 1] values # npy_uint8
where = numpy.flatnonzero
n = len(x)
starts = numpy.r_[0, where(~numpy.isclose(x[1:], x[:-1], equal_nan=True)) + 1]
lengths = numpy.diff(numpy.r_[starts, n])
values = x[starts]
# Reduce repetitions of color to max 0x7D/125 and store in bytearray
rleData = bytearray()
cdef unsigned int nr=0
cdef unsigned int col=0
cdef unsigned char color=0
cdef unsigned char encValue = 0
cdef unsigned int l=len(lengths)
cdef unsigned int i=0
for i in range (0,l):
nr=lengths[i]
col=values[i]
# color = (abs(col)>1) # slow
color = 1 if col else 0 # fast
while nr > 0x7D:
encValue = (color << 7) | 0x7D
rleData.append(encValue)
nr = nr - 0x7D
encValue = (color << 7) | nr
rleData.append(encValue)
# Needed is an byte string, so convert
return bytes(rleData)
我编译RLE编码.pyx与python setup.py build_ext --inplace
一起使用设置.py. 你知道吗
from distutils.core import setup
from Cython.Build import cythonize
import numpy
setup(
ext_modules=cythonize("rleEncode.pyx"),
include_dirs=[numpy.get_include()]
)
(我使用命令sudo apt-get install python3-dev
来安装所需的python.h头)
用你的评论和https://kogs-www.informatik.uni-hamburg.de/~seppke/content/teaching/wise1314/20131128_letsch-gries-boomgarten-cython.pdf,我的速度快了10倍!你知道吗
生成的代码是:
相关问题 更多 >
编程相关推荐