Convolve2d just by using Numpy
You could generate the subarrays using as_strided
:
import numpy as np
a = np.array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14],
[15, 16, 17, 18, 19],
[20, 21, 22, 23, 24]])
sub_shape = (3,3)
view_shape = tuple(np.subtract(a.shape, sub_shape) + 1) + sub_shape
strides = a.strides + a.strides
sub_matrices = np.lib.stride_tricks.as_strided(a,view_shape,strides)
To get rid of your second "ugly" sum, alter your einsum
so that the output array only has j
and k
. This implies your second summation.
conv_filter = np.array([[0,-1,0],[-1,5,-1],[0,-1,0]])
m = np.einsum('ij,ijkl->kl',conv_filter,sub_matrices)
# [[ 6 7 8]
# [11 12 13]
# [16 17 18]]
Strided convolution of 2D in numpy
Ignoring the padding argument and trailing windows that won't have enough lengths for convolution against the second array, here's one way with np.lib.stride_tricks.as_strided
-
def strided4D(arr,arr2,s):
strided = np.lib.stride_tricks.as_strided
s0,s1 = arr.strides
m1,n1 = arr.shape
m2,n2 = arr2.shape
out_shp = (1+(m1-m2)//s, m2, 1+(n1-n2)//s, n2)
return strided(arr, shape=out_shp, strides=(s*s0,s*s1,s0,s1))
def stride_conv_strided(arr,arr2,s):
arr4D = strided4D(arr,arr2,s=s)
return np.tensordot(arr4D, arr2, axes=((2,3),(0,1)))
Alternatively, we can use the scikit-image built-in view_as_windows
to get those windows elegantly, like so -
from skimage.util.shape import view_as_windows
def strided4D_v2(arr,arr2,s):
return view_as_windows(arr, arr2.shape, step=s)
Batch convolution 2d in numpy without scipy?
Just need to get shape
to be 5d and get the strides
to match the shape
.
shape = f.shape + (x.shape[0],) + tuple(np.subtract(x.shape[1:], f.shape) + 1)
strides = (x.strides * 2)[1:]
M = np.lib.stride_tricks.as_strided(x, shape=shape, strides=strides)
y = np.einsum('pq,pqbmn->bmn', f, M)
now M
might get really big if b
gets really big, but it works on your toy problem.
Fastest 2D convolution or image filter in Python
It really depends on what you want to do... A lot of the time, you don't need a fully generic (read: slower) 2D convolution... (i.e. If the filter is separable, you use two 1D convolutions instead... This is why the various scipy.ndimage.gaussian
, scipy.ndimage.uniform
, are much faster than the same thing implemented as a generic n-D convolutions.)
At any rate, as a point of comparison:
t = timeit.timeit(stmt='ndimage.convolve(x, y, output=x)', number=1,
setup="""
import numpy as np
from scipy import ndimage
x = np.random.random((2048, 2048)).astype(np.float32)
y = np.random.random((32, 32)).astype(np.float32)
""")
print t
This takes 6.9 sec on my machine...
Compare this with fftconvolve
t = timeit.timeit(stmt="signal.fftconvolve(x, y, mode='same')", number=1,
setup="""
import numpy as np
from scipy import signal
x = np.random.random((2048, 2048)).astype(np.float32)
y = np.random.random((32, 32)).astype(np.float32)
""")
print t
This takes about 10.8 secs. However, with different input sizes, using fft's to do a convolution can be considerably faster (Though I can't seem to come up with a good example, at the moment...).
Related Topics
How to Specify an Authenticated Proxy for a Python Http Connection
Replacing Text in a File with Python
How to Get Last Items of a List in Python
Heapq with Custom Compare Predicate
How to Use Pil to Make All White Pixels Transparent
Get Raw Post Body in Python Flask Regardless of Content-Type Header
Python Datetime Object Show Wrong Timezone Offset
Why Don't Methods Have Reference Equality
Python's JSON Module, Converts Int Dictionary Keys to Strings
Sqlalchemy: What's the Difference Between Flush() and Commit()
Python Pandas Extract Year from Datetime: Df['Year'] = Df['Date'].Year Is Not Working
Use Pytesseract Ocr to Recognize Text from an Image
Debugging (Displaying) SQL Command Sent to the Db by SQLalchemy
What Do I Use for a Max-Heap Implementation in Python
Configuring Spark to Work with Jupyter Notebook and Anaconda