lzqsd / inverserenderingofindoorscene Goto Github PK

Hi, This is Prof. Moon R. Jung from Sogang University, Seoul, Korea.
I recently read your paper with a great interest. I feel sorry to ask these too simple questions,
especially as a person with a Ph.D. in computer graphics.

When I first read the paper, of course I thought I knew what you meant by these terms.
But then I read the paper you referred to in the paper:

Real Shading in Unreal Engine 4 by Brian Karis, Epic Game:
https://blog.selfshadow.com/publications/s2013-shading-course/karis/s2013_pbs_epic_notes_v2.pdf

This paper talks about the so-called "Metallic-Roughness workflow" and talks about
material model:
BaseColor,
Metalic,
Roughness.

After it, I read the code of OptixRenderer, in particular shade code microfacet.cu.

I found out that it uses the Metalic workflow. The following line strongly suggests it:

float3 fresnel = F0 * (1 - metallicValue) + metallicValue * albedoValue; // MJ: fresnel = SpecularMap

albedoValue = (1 - metallicValue) * albedoValue; //MJ: albedoValue on on the right hand side is BaseColorMap

//MJ ... is my comment.

Q1. Is your "diffuse albedo" actually BaseColorMap?

Q2. The metalic workflow talks about "roughness". Is your "specular roughness" the same as "roughness"?

Hi, may I get the access of download link here? I have sent email to [email protected], but do not get the response. My email is [email protected]. Waiting for your reply, thanks.

Hi @lzqsd
Very amazing work which inspires me quite a lot! But I still have several questions, and I hope you could kindly clarify them.

1. The x and y axis of the coordinate system of hemisphere used in illumination. Why use L1 (p=1) normalization instead of L2, and, why get camx by reversing the cross product of camy and normalPred?
   

   InverseRenderingOfIndoorScene/models.py

   Line 480 in eeac1e9

   camx = -F.normalize(torch.cross(camy, normalPred,dim=1), p=1, dim=1)

2. I have also a general question about the linear space and gamma space. The input image I is gamma-corrected LDR, and the output of the render equation is the radiance in linear space. Thus I am wondering whether it makes sense to compute the render loss, although with a scale factor, as shown in Eqn. (7) in the paper?

Looking forward to your reply, and thanks a lot in advance!

Hi, I encountered the following problems

this problem is caused by this

I have downloaded OpenRooms dataset and put them in the ./dataset folder, there are no pictures of png format in main_xml1 folder. Now, I don't know where are the pictures of png format from.

Hi, may I get the access of download link here? I have sent email to [email protected] host, but do not get the response. My email is [email protected]. Waitting for your reply, thanks.

Hi guys,

Do you have any estimation about when you will be able to release the openrooms dataset and tools?

Thank you.

Hello,
I have tried to test the model on a single gpu and windows. Unfortunately, I faced too many errors.
My current problem is the following:
The "freeze_support()" line can be omitted if the program
is not going to be frozen to produce an executable.

I solved the problem according to this link.
But, now when I run the code it quickly stops without producing outputs.

Has anyone ran this repo on a single GPU?

Hi! Have you tried the training with higher resolutin like 720*1080? I saw you mentioned networks can generate smoother results with smaller resolution, and if the result will deteriorate a lot with high resolution?

Thanks for your great work! And I was wondering if it is possible that you could provide the code for inferencing on a customized single indoor scene image especially for BRDF estimation?

Hi,sorry for disturbing you.
I got the download link, But I can’t open a Google file or folder.It says:"The organization that owns this content does not allow you to access it".
Google drive give me this Tip: If you need access straight away, contact the owner of the file and ask them to give you access.

Here's my email: [email protected].
I have sent you an email for this problem.
Looking forward to your reply:)

Hi!

How hard do you think it would be to translate the Optix Renderer scene description files to Mitsuba scene description files?

I think it would be interesting to be able to render this dataset using Mitsuba and some of the extensions available for Mitsuba such as MitsubaToF (for transient rendering) and MitsubaCLT (for computational light transport imaging systems), and maybe even Mitsuba 2 (for differentiable rendering).

Thanks!
Felipe

Hi, which file is the material editing code in? I am interested in material editing and want to study it.

@lzqsd I download the pretrained model and use testReal.py to render img, but the result is very blur(input img is clear)

It seems the Model Download link is unable to access, could you please confirm? Thanks!

Hi, thank you for your good work.

I was wondering how I can extract the Diffuse Albedo, Normal, and Roughness that's tileable out of the model after running it? I tried running the model, but the output images are not tileable.

Thanks

I do not know whether you have considered to functionality of loading IES files for the corresponding light sources in the scene, but I believe that this would be quite a nice addition.

Thanks.

Hello, I'm very interested in your work. But i have one question. How to estimate lighting information from a limited range LDR image without any other information? testReal.py seems that it can only output the material estimate of the scene.