About metrics update about sparse-depth-completion HOT 4 CLOSED

Hi @longyangqi,

For the correct calculation, you should take into account that the ground truth is semi-dense. This means that we need to keep track of the valid points per image. Take a look at the evaluation script of KITTI to see the detailed calculation of the RMSE for the benchmark.

Best,
Wouter

from sparse-depth-completion.

Hi @longyangqi,

For the correct calculation, you should take into account that the ground truth is semi-dense. This means that we need to keep track of the valid points per image. Take a look at the evaluation script of KITTI to see the detailed calculation of the RMSE for the benchmark.

Best,
Wouter

Thank you for your reply!
Yeah, I did not notice the evaluation script of KITTI. I have read it now !
Your implementation is corresponding to the function statWeightedMean in the evaluation script of KITTI cpp/utils.h. Right?
But the eval function using 'statMean' instead of 'statWeightedMean' ?
I'm still confused about that.

Thanks!

from sparse-depth-completion.

Hi @longyangqi,

You could be right. My calculation is only an approximation since I calculate the RMSE on the whole batch and not per image as KITTI and I do a weighing dependent on the amount of points. I don't think it matters much to be honest as long as the comparisons are consistent.

Regards,
Wouter

from sparse-depth-completion.

I get it.

Thanks a lot!

from sparse-depth-completion.