open-mmlab/mmdetection

why need to add 0.5

SparKgod1 opened this issue · 1 comments

https://mmdetection.readthedocs.io/en/latest/user_guides/train.html#train-with-customized-datasets:~:text=poly%20%3D%20%5B(x%20%2B%200.5%2C%20y%20%2B%200.5)%20for%20x%2C%20y%20in%20zip(px%2C%20py)%5D
Why this example of converting data format to 'coco' need to add 0.5?
Does it affect my training if I ignore this?
looking forward to your reply~

here

  def convert_balloon_to_coco(ann_file, out_file, image_prefix):
  ...
  poly = [(x + 0.5, y + 0.5) for x, y in zip(px, py)]