JiawangBian / SC-SfMLearner-Release

Unsupervised Scale-consistent Depth Learning from Video (IJCV2021 & NeurIPS 2019)
https://jwbian.net/sc-sfmlearner
GNU General Public License v3.0
733 stars 149 forks source link

About depth #33

Open TopGun666 opened 4 years ago

TopGun666 commented 4 years ago

Hi @JiawangBian

I have a question about depth interpolation.

Here is the statement in your paper. You get the warped $D^{a}{b}$ by converting $D{a}$ a to 3D space and projecting to the image plane of $I{b}$.$D'{b}$ is interpolated from $D{b}$. $D{b} is the estimated depth maps of $I{b}$. And you also implement this in your code. Note that we cannot directly use $D{b}$ because the warping flow does not lie on the pixel grid.

In my mind, the depth in $D{b}$ corresponds to each pixel of the original image. It does lie on the pixel grid. It does not require interpolation. In contrast, $D^{a}{b}$ does not lie on the pixel grid because it is warpped from $D{a}$. I think $D^{a}{b}$ the one who needs interpolation.

After I read this paper by SfMLearner, I had such doubts. I wonder if this is the place where your thesis is inconsistent with the original. Looking forward to your opinion.

Regards,

Yu

livey commented 2 years ago

I have the same concern. Have you figured it out?

lamduykhang4869 commented 2 years ago

I have the same concern. Have you found the solution? Can we discuss more about this problem?