Wrong coordinates of Darknet Annotations (CLIPPED)

[Deleted user]

I have downloaded your data set Darknet Annotations (CLIPPED), but I found that the coordinates of cx, cy are wrong, and the label value is only 1/2 of the real value, and rely on the following script to convert:

import os
dirs = [
    "/path/to/train2017", "/path/to/val2017"
]
for di in dirs:
    labels = os.listdir(di)
    for label_file in labels:
        lb_path = os.path.join(di, label_file)
        with open(lb_path, "r") as f:
            labels = f.readlines()
        new_labels = []
        for label in labels:
            id, cx, cy, w, h = map(float, label.split())
            id = str(int(id))
            x = 2*(cx-w/2)
            y = 2*(cy-h/2)
            ncx = x+w/2
            ncy = y+h/2
            vals = [id, str(ncx), str(ncy), str(w), str(h)]
            new_labels.append(" ".join(vals)+"\n")
        with open(lb_path, "w") as f:
            f.writelines(new_labels)