Upgrade OpenCV version #557
Conversation
| @@ -162,7 +162,7 @@ def _draw_tag( | |||
| ) | |||
| bbox_width = btm_right[0] - top_left[0] | |||
| offset = int((bbox_width - text_width) / 2) | |||
| position = (int(top_left[0] + offset), int(top_left[1] - baseline)) | |||
| position = (top_left[0] + offset, int(top_left[1] - baseline)) | |||
There was a problem hiding this comment.
cv2.getTextSize() returns (text_width, text_height), baseline, all of which are integers.
So there is no need to int() once the coords are integers.
The same applies to line 111 in bbox.py for _draw_label().
Verified from opencv C++ docs, source, and runtime printouts:
*getTextSize: text_width=<class 'int'>, text_height=<class 'int'>, baseline=<class 'int'>
There was a problem hiding this comment.
Wasn't able to find this, thanks! Updated.
| @@ -59,4 +59,4 @@ def project_points_onto_original_image( | |||
| projected_points[:, 0] *= width | |||
| projected_points[:, 1] *= height | |||
|
|
|||
| return projected_points | |||
| return projected_points.astype(int) | |||
There was a problem hiding this comment.
Just a note that numpy.astype() does not round up the numbers:
>>> import numpy as np
>>> x = np.array([1.1, 2.5, 3.7])
>>> x
array([1.1, 2.5, 3.7])
>>> x.astype(int)
array([1, 2, 3])
If intention is to round, can do this:
>>> np.round(x).astype(int)
array([1, 2, 4])
NB: numpy implements the round-half-to-even method.
There was a problem hiding this comment.
Implemented rounding for that extra bit of accuracy.
Closes #556.
4.5.2.54chosen due to factors mentioned in the issue.