Question about visualization of the correlation

[pyni]

Hi, I wonder why there is no data in the red parts of the figure: And here angle difference denotes the rotation error between estimated rotation and ground truth rotation, right? Ideally, for different test images, if they have the same rotation errors, their representation distances should be the same, am I right? And may I know if the representation distances in your paper use the 2D binary visibility mask?

Thanks.

[nv-nguyen]

Hi, Very good remarks! I am not sure neither. As in other quantitative and qualitative evaluations, I always use 2D binary mask of templates but do not use visible mask of query images. Please note that using masks only changes "descriptor diff" in Y axis, not "angle diff" in X axis for data points. I will let you know if I find the reason why there is no data points in this region.

[pyni]

OK， i got it. Thanks.