Giters

lllyasviel / AppearanceEraser

Erasing Appearance Preservation in Optimization-based Smoothing (ECCV 2020)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

About the "EDL1" in your paper

int2char opened this issue · comments

commented

Is the code of EDL1 or the supplemental material mentioned in the paper which described the detail of EDL1 available?

commented

EDL1 = L1 + EAP automatic erasing + DPGMM

"L1 + EAP automatic erasing" is the "L1 smoothing with EAP automatic erasing" in this project.

DPGMM can be found in the original "L1 smoothing" paper:

"An L1 Image Transform for Edge-Preserving Smoothing and Scene-Level Intrinsic Decomposition"

http://cseweb.ucsd.edu/~bisai/papers/SIGGRAPH15_IntrinsicDecomposition.pdf

Page 6, Section 4.1: "Clustering"

image

Please refer to their implementation details for the DPGMM application.

We do not have plan to upload the DPGMM codes because it belongs to third party and I do not have the right to opensource it.

If you want the DPGMM code, you may directly contact the original "L1 smoothing" authors of Sai Bi, Xiaoguang Han, and Yizhou Yu.

commented

Thanks for the detailed interpretation，I have an another question: how can I separate the specular reflection layer and the shadow layer of an image as what is shown in your project page? I read the paper "Illumination Decomposition for Material Recoloring with Consistent Interreflections" as you cited which supports assigning specular reflections into seperateted layers, but I failed to find the correlated method. Is there any redesigning of the energy functions ?

commented

No you do not need to modify the energy function. The specular reflection layer can be directly computed by inverting the shadow map.

Assuming you have albedo A and shadow S and you decompose the image I with "Illumination Decomposition for Material Recoloring with Consistent Interreflections" (IDMRCI) as

I = A * S

Then you can use a simple inverting operation "F(x)=1.0-x " and then

F(I) = F(A) * F(S)

Then IDMRCI will decompose specular reflection and S is the specular reflection.

I believe I find this trick in IDMRCI but currently I cannot find the related descriptions in that paper. Later I will take another look at it.

Also if you are interested in those transform, you may also want this paper with many related transforms:

https://koyama.xyz/project/color_unblending/
commented

thanks, this really help a lot