ubc-vision/image-matching-benchmark

[question] keypoints coordinate convention

Closed this issue · 4 comments

santelelle commented

In the benchmark submit section it is stated:

The keypoint file must contain an array with N keypoints with 0-indexed (x, y) coordinates, with the origin in the top left corner.

Does this mean that the top-left corner of the top-left pixel has coordinates (0, 0) and the center of this same pixel is at (0.5, 0.5)?

ducha-aiki commented

No. It just means, not matlab 1-based indexing :)

santelelle commented

Ok thanks :)
So which convention should I use when submitting keypoints coordinate?

  • A: the top-left corner of the top-left pixel has coordinate (x, y)=(0, 0), therefore the center of the same pixel has coordinate (0.5, 0.5) --> ETH3D/Colmap convention

  • B: the top-left corner of the top-left pixel has coordinate (x, y)=(-0.5, -0.5), therefore the center of the same pixel has coordinate (0, 0) --> OpenCV convention

ducha-aiki commented

Top-left at (0,0)

santelelle commented

Ok thanks, so to close this the right convention is the A