Confusion about the inverse-depth-smooth-loss

[SJoJoK]

Hello! I'm currently doing some research on NeRF, and I found your work SinNeRF.
It's an awesome work, and I download the code (in this repository).
While reading the code, I found that you use kornia.losses.inverse_depth_smoothness_loss to calculate the "self-supervised inverse depth smoothness loss" which you describe in the paper (equation 4) .
The conflict is that, while kornia.losses.inverse_depth_smoothness_loss using the first-order gradient of the RGB Image according to the official document and its source code, the equation in paper use the second-order gradient of RGB Image.
Thanks for any advice and help.

[Ir1d]

Hi @SJoJoK, Good catch! Equation 4 is indeed confusing. I will update the draft to align with the code.
Previously we used some more complex depth smoothness loss, and later found it easier to use the kornia implementation.
Also note that for the latest implementation, we observe that sometimes disabling the depth smoothness loss can lead to a better result.
Thank you again!

[SJoJoK]

Got it, thanks for the information : )