Python 创建三维二值图像
我有一个二维数组,Python 创建三维二值图像,python,numpy,indexing,binary,ndimage,Python,Numpy,Indexing,Binary,Ndimage,我有一个二维数组,a,包含一组100 x,y,z坐标: [[ 0.81 0.23 0.52] [ 0.63 0.45 0.13] ... [ 0.51 0.41 0.65]] 我想创建一个三维二值图像,b,在x、y、z维度上各有101个像素,坐标范围在0.00和1.00之间。 由a定义的位置处的像素值应为1,所有其他像素值应为0 我可以用b=np.zeros((101101101))创建一个正确形状的零数组,但是我如何分配坐标并将其切片以使用a创建这些零?您可以这样做- #
a
,包含一组100 x,y,z坐标:
[[ 0.81 0.23 0.52]
[ 0.63 0.45 0.13]
...
[ 0.51 0.41 0.65]]
我想创建一个三维二值图像,b
,在x、y、z维度上各有101个像素,坐标范围在0.00和1.00之间。
由a
定义的位置处的像素值应为1,所有其他像素值应为0
我可以用
b=np.zeros((101101101))
创建一个正确形状的零数组,但是我如何分配坐标并将其切片以使用a
创建这些零?您可以这样做-
# Get the XYZ indices
idx = np.round(100 * a).astype(int)
# Initialize o/p array
b = np.zeros((101,101,101))
# Assign into o/p array based on linear index equivalents from indices array
np.put(b,np.ravel_multi_index(idx.T,b.shape),1)
分配部分的运行时- 让我们使用更大的网格来计时
In [82]: # Setup input and get indices array
...: a = np.random.randint(0,401,(100000,3))/400.0
...: idx = np.round(400 * a).astype(int)
...:
In [83]: b = np.zeros((401,401,401))
In [84]: %timeit b[list(idx.T)] = 1 #@Praveen soln
The slowest run took 42.16 times longer than the fastest. This could mean that an intermediate result is being cached.
1 loop, best of 3: 6.28 ms per loop
In [85]: b = np.zeros((401,401,401))
In [86]: %timeit np.put(b,np.ravel_multi_index(idx.T,b.shape),1) # From this post
The slowest run took 45.34 times longer than the fastest. This could mean that an intermediate result is being cached.
1 loop, best of 3: 5.71 ms per loop
In [87]: b = np.zeros((401,401,401))
In [88]: %timeit b[idx[:,0],idx[:,1],idx[:,2]] = 1 #Subscripted indexing
The slowest run took 40.48 times longer than the fastest. This could mean that an intermediate result is being cached.
1 loop, best of 3: 6.38 ms per loop
首先,从安全地将浮点数舍入整数开始。在上下文中,请参见问题
a_indices = np.rint(a * 100).astype(int)
接下来,将b
中的索引分配给1。但是要小心使用普通的列表
而不是数组,否则会触发使用。这种方法的性能似乎与其他方法的性能相当(感谢@Divakar!:-)
我创建了一个小示例,大小为10而不是100,尺寸为2而不是3,以说明:
>>> a = np.array([[0.8, 0.2], [0.6, 0.4], [0.5, 0.6]])
>>> a_indices = np.rint(a * 10).astype(int)
>>> b = np.zeros((10, 10))
>>> b[list(a_indices.T)] = 1
>>> print(b)
[[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 1. 0. 0. 0.]
[ 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 1. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]]
(a*100)。aType(int)
非常危险!你可能会得到错误的索引!试试看:a=np.arange(101)/100
,b=(100*a).astype(int)
,(b==np.arange(101,dtype=int)).all()
,你会得到False
!例如,由于浮点精度问题,数字28在b
中出现两次。@Praveen Damn,你说得对!四舍五入是唯一的办法。更新了帖子。现在,它和你开始时的一样。你想让我删除这篇文章吗?我不会要求你删除它。你的np.put
方法很有趣。我对这两者之间的时间分析很感兴趣,只要使用list(a_index.T)
@Praveen当然,让我补充一下。@Praveen补充道。性能数据似乎是可比的。
>>> a = np.array([[0.8, 0.2], [0.6, 0.4], [0.5, 0.6]])
>>> a_indices = np.rint(a * 10).astype(int)
>>> b = np.zeros((10, 10))
>>> b[list(a_indices.T)] = 1
>>> print(b)
[[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 1. 0. 0. 0.]
[ 0. 0. 0. 0. 1. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 1. 0. 0. 0. 0. 0. 0. 0.]
[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.]]