lip aspect ratio calculation issue

I am trying to assess whether the lips of a person are moving too much while the mouth is closed (to conclude it is chewing).

The mouth closed part is done without any issue, but when I try to assess the lip movement through landmarks (dlib) there seems to be a problem with the last landmark of the mouth.

Inspired by the mouth example (https://github.com/mauckc/mouth-open/blob/master/detect_open_mouth.py#L17), I wrote the following function:

def lips_aspect_ratio(shape):
    # grab the indexes of the facial landmarks for the lip
    (mStart, mEnd) = (61, 68)
    lip = shape[mStart:mEnd]
    print(len(lip))
    # compute the euclidean distances between the two sets of
    # vertical lip landmarks (x, y)-coordinates
    # to reach landmark 68 I need to get lib[7] not lip[6] (while I get lip[7] I get IndexOutOfBoundError)
    A = dist.euclidean(lip[1], lip[6])  # 62, 68
    B = dist.euclidean(lip[3], lip[5])  # 64, 66

    # compute the euclidean distance between the horizontal
    # lip landmark (x, y)-coordinates
    C = dist.euclidean(lip[0], lip[4])  # 61, 65

    # compute the lip aspect ratio
    mar = (A + B) / (2.0 * C)

    # return the lip aspect ratio
    return mar

The landmark of the lips are (61, 68), when I extract the lip as lip = shape[61:68] and try to access the last landmark as lip[7] I get the following error:

IndexError: index 7 is out of bounds for axis 0 with size 7

Why is that? and How to get the last landmark of the lip/face
Otherwise How to define an aspect ratio for the lips to conclude they are moving significantly?

>Solution :

lip = shape[61:68]

The slices exclude the end element. So you got 7 elements: 61,62,63,64, 65,66,67. And len(lip) == 7 confirms that.

If there truly are 8 points for the lip shape, and they include element 68, the slice should be:

lip = shape[61:69]
assert(len(range(61, 69) == 8)
assert(len(lip) == 8)

Note that if shape is sufficiently long, then len(shape[a:b:c]) == len(range(a, b, c)). That’s because the range function acts like taking a slice out of an infinitely long list of integers (with caveats).

So, the problem is a classic off-by-one, not much to do with AI/image analysis 🙂