What is this Python syntax? rect[:, 2:] += rects[:, 2]

Question

In BlackHatPython, Chapter 4 has this function:

def face_detect(path,file_name):
  img = cv2.imread(path)
  cascade = cv2.CascadeClassifier("haarcascade_frontalface_alt.xml")
  rects = cascade.detectMultiScale(img, 1.3, 4, cv2.cv.CV_HAAR_SCALE_IMAGE, (20,20))
  if len(rects) == 0:
    return False
  rects[:, 2:] += rects[:, :2]
  # highlight the faces in the image
  for x1,y1,x2,y2 in rects:
    cv2.rectangle(img,(x1,y1),(x2,y2),(127,255,0),2)
  cv2.imwrite("%s/%s-%s" % (faces_directory,pcap_file,file_name),img)
  return True

What is the meaning of

rect[:, 2:] += rects[:, :2]

looks like `numpy` to me. https://docs.scipy.org/doc/numpy/reference/arrays.indexing.html — Patrick Haugh, Nov 08 '16 at 14:06
@Farhan.K A slice can consist of a comma-separated sequence of *proper* slices, which have the general form `start:step:stop`. Built-in types only accept proper slices; the full form was added to the language for `numpy`. Any user-defined class, though, can accept them. — chepner, Nov 08 '16 at 14:10
With `rects[:, :2]`, the argument to `type(rects).__getitem__()` would be a tuple of `slice` objects: `((slice(None, None, None), slice(None, 2, None))`. — chepner, Nov 08 '16 at 14:12
@Farhan.K since you get a `TypeError` it has passed the parsing, that is it's valid syntax. Then `m` of course must be of a type that supports the operation. — skyking, Nov 08 '16 at 14:16
This question is not really a duplicate since it is not just about the syntax but is about what this line of code really does... The answer is that it converts an array of rects like [x, y, w, h] to an array of [x, y, x+w, y+h], which is easier to use afterwards. — Vinicius, Jun 15 '18 at 22:53

score 4 · Accepted Answer · answered Nov 08 '16 at 14:07

This is slicing arrays in python, as e.g. explained here: Explain Python's slice notation

In you specific case, you have a numpy array, which is a 2D structure, so there are two dimensions on which you can slice. The individual slices are separated by the comma. Look at this code for a visualization:

In [7]: np.diag([3,4,5])
Out[7]: 
array([[3, 0, 0],
       [0, 4, 0],
       [0, 0, 5]])

In [8]: np.diag([3,4,5])[:2]
Out[8]: 
array([[3, 0, 0],
       [0, 4, 0]])

In [9]: np.diag([3,4,5])[:2, :]
Out[9]: 
array([[3, 0, 0],
       [0, 4, 0]])

In [10]: np.diag([3,4,5])[:2, :1]
Out[10]: 
array([[3],
       [0]])

score 0 · Answer 2 · answered Nov 08 '16 at 14:17

To dissect what the expression is you have to consult the Python Language Reference. What it is is that the subexpressions : and 2: are slicings, then the expression :, 2: is two slicings joined by the comma operator (creating a tuple of two slicings). Then that is used to subscribe rect and rects and whatever that results in will be applied to the += operator.

What is this Python syntax? rect[:, 2:] += rects[:, 2]

2 Answers2