是否有“边界框"?NumPy 中 ndarray 的函数(具有非零值的切片)? [英] Is there a "bounding box" function (slice with non-zero values) for a ndarray in NumPy?

问题描述

我正在处理通过 numpy.array() 创建的数组，我需要在模拟图像的画布上绘制点.由于包含有意义数据的数组中心部分周围有很多零值，我想修剪"数组，删除只包含零的列和只包含零的行.

I am dealing with arrays created via numpy.array(), and I need to draw points on a canvas simulating an image. Since there is a lot of zero values around the central part of the array which contains the meaningful data, I would like to "trim" the array, erasing columns that only contain zeros and rows that only contain zeros.

所以，我想知道一些原生的 numpy 函数，甚至是一个代码片段来修剪"或找到一个边界框"来只对数组中包含数据的部分进行切片.

So, I would like to know of some native numpy function or even a code snippet to "trim" or find a "bounding box" to slice only the data-containing part of the array.

(因为这是一个概念性的问题，我没有放任何代码，对不起，如果我应该，我很新鲜在 SO 上发帖.)

(since it is a conceptual question, I did not put any code, sorry if I should, I'm very fresh to posting at SO.)

感谢阅读

推荐答案

下面的代码，来自 这个答案在我的测试中运行得最快:

The code below, from this answer runs fastest in my tests:

def bbox2(img):
    rows = np.any(img, axis=1)
    cols = np.any(img, axis=0)
    ymin, ymax = np.where(rows)[0][[0, -1]]
    xmin, xmax = np.where(cols)[0][[0, -1]]
    return img[ymin:ymax+1, xmin:xmax+1]

使用 argwhere 接受的答案有效但运行速度较慢.我的猜测是，这是因为 argwhere 分配了一个巨大的索引输出数组.我在一个大型 2D 阵列(一个 1024 x 1024 的图像，大约有一个 50x100 的非零区域)上进行了测试.

The accepted answer using argwhere worked but ran slower. My guess is, it's because argwhere allocates a giant output array of indices. I tested on a large 2D array (a 1024 x 1024 image, with roughly a 50x100 nonzero region).

这篇关于是否有“边界框"?NumPy 中 ndarray 的函数(具有非零值的切片)?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！