MoGe icon indicating copy to clipboard operation
MoGe copied to clipboard

Question about the evaluation of point map estimation.

Open chenhaomingbob opened this issue 1 year ago • 1 comments

Thank you for your interesting work.

I am not familiar with the direction of point map estimation. I have read Section 4.1 in your paper for quantitative evaluations. But I still do not understand how to align the predicted point map with the GT point map before calculating the point error.

chenhaomingbob avatar Nov 07 '24 08:11 chenhaomingbob

Hi. The scale (and shift) of prediction $\{\hat p_i\}$ are aligned with the ground truth $\{p_i\}$ as follows.

  • For evaluation of scale-invariant point map,

$$ s^*=\mathop{argmin}_s \sum_i {1\over z_i}\| s\hat p_i-p_i\| $$

  • For evaluation of affine-invariant point map,

$$ (s^*,t^*)=\mathop{argmin}_{s,t} \sum_i {1\over z_i}\| s\hat p_i+t-p_i\| $$

They are solved with our proposed solver (without clipping). See section A.2 for details of the algorithm.

EasternJournalist avatar Nov 11 '24 13:11 EasternJournalist