`[0, 267, 270, 468]` describes a bbox, how do I get it from `[266.67, 0.0, 201.69, 269.58]`?

Question

i got a set of modified annotations for a bunch of coco images. for example, [0, 267, 270, 468] and [254, 250, 458, 454] are 2 pieces from the set, describing two bboxes for the following image. (named 000000173350.jpg in the original dataset)

although they are not in the form of [x, y, width, height], the following ones, which come from the original dataset, are

[266.67, 0.0, 201.69, 269.58] and [250.18, 254.3, 203.64, 203.64]

with the original annotations i can draw bboxes easily

i can decode some part of the modified annotations, since the original ones could be rephrased as [267, 0, 202, 270] (ceiling) and [250, 254, 203, 203] (floor), and the xs and ys parts are swapped.

however, i cannot imagine the rest of the modified annotations, how do i get them from the original annotations?

Answer 1

If I understand correctly you want to achieve the [x0, y0, x1, y1] format, denoting the top-left (x0, y0) and bottom-right (x1, y1) coordinates from the [y, x, h, w] format you have. If so, you can do as follows:

def convert_bbox(box: tuple) -> tuple:
    y0, x0, h, w = box

    x0 = round(x0)
    y0 = round(y0)
    x1 = x0 + round(w)
    y1 = y0 + round(h)

    return x0, y0, x1, y1

which yields:

>>> convert_bbox([266.67, 0.0, 201.69, 269.58])
(0, 267, 270, 469)

>>> convert_bbox([250.18, 254.3, 203.64, 203.64])
(254, 250, 458, 454)

Answer 2

I rounded your values and turned them into variables to show how your lists are build up. With the forumlas inside my example lists you should be able to find out how certain values came up.

a=267
b=0
c=202
d=270

List2= [a,b,c,d] #267,0,202,270
List1= [b,a,b+d,a+c] #0,267,0+270,267+202 -> 0, 267,270,469