Visualization of LUTs

[huizeng]

Visualization of LUTs

[hkzhang95]

Hi, I write a simple python script for visualization at #37. Please have a check about whether it is correct.

[HuiZeng]

Hi, sorry for the late reply.
I do have a matlab code to visualize the learned 3D LUTs. However, I am currently taking a quarantine and unavilable to this code.
I will update the code after the quarantine.

[tnt-ooo-tnt]

I am so afraid to hear that
I hope you get well soon, cheers 👍

[HuiZeng]

Hi， I have uploaded the code for visualization of learned luts.

There are two steps:

1. run the python code to write the learned lut into a .txt file.
2. run the matlab code to visualize the 3d lut.

You may need to check the paths and file names according to your own enviroments.

[hkzhang95]

I am a little confused about the visualization results.

From my understanding, the visualization of 3D LUT should be a point cube in which each point matches a new color. Here is an example from the internet:

However, the visualization results from Fig.4 in the paper are not exact cubes. So I am wondering about the meanings of the axes in each figure. Maybe for each component (e.g. R), the visualization should be a cube, then each point (R_in, G_in, B_in) matches a new R-value?

[HuiZeng]

Hi， the example you provide is indeed more vivid. In this 3D cube, each point contains three values, namely, the {R_out, G_out. B_out}, rather than one single value. It actually contains 4D information. Unfortunately, I do not know how to plot 3D LUT like this. Also it is inconvenient to see the details in such a dense plot.

Instead, I choose to visualize the R, G, B channel independently. Your understanding is right. The three axes represent the R_in, G_in, B_in while the output R (or G, or B) values are plotted as multiple slices rather than one cube. You can also choose to visualize all R values as a cube if you like.

[tnt-ooo-tnt]

Thank you for your effort a lot,

I will let this issue be opened for a several days if you don't mind for derived discussion.

Thank you again so much

[hkzhang95]

Hi， the example you provide is indeed more vivid. In this 3D cube, each point contains three values, namely, the {R_out, G_out. B_out}, rather than one single value. It actually contains 4D information. Unfortunately, I do not know how to plot 3D LUT like this. Also it is inconvenient to see the details in such a dense plot.

Instead, I choose to visualize the R, G, B channel independently. Your understanding is right. The three axes represent the R_in, G_in, B_in while the output R (or G, or B) values are plotted as multiple slices rather than one cube. You can also choose to visualize all R values as a cube if you like.

Maybe I did not make myself clear enough.

You said that The three axes represent the R_in, G_in, B_in. But from your code, the third axis represents the value of R/G/B. So from my viewpoint, one of the axis information is missing, and this is the reason why the visualization result is not a standard cube.

[hkzhang95]

I will provide a python script to visualize the LUT like the example I provided if I am available. : ）

[HuiZeng]

Hi， the example you provide is indeed more vivid. In this 3D cube, each point contains three values, namely, the {R_out, G_out. B_out}, rather than one single value. It actually contains 4D information. Unfortunately, I do not know how to plot 3D LUT like this. Also it is inconvenient to see the details in such a dense plot.
Instead, I choose to visualize the R, G, B channel independently. Your understanding is right. The three axes represent the R_in, G_in, B_in while the output R (or G, or B) values are plotted as multiple slices rather than one cube. You can also choose to visualize all R values as a cube if you like.

Maybe I did not make myself clear enough.

You said that The three axes represent the R_in, G_in, B_in. But from your code, the third axis represents the value of R/G/B. So from my viewpoint, one of the axis information is missing, and this is the reason why the visualization result is not a standard cube.

Yes, you are right. I just checked the code. It actually plots (R_in, G_in, R_out/G_out/B_out) and there are a total of six dimensions (R_in, G_in, B_in, R_out, G_out, B_out).
That will be great if you can provide a more elegent visualization script.

[HuiZeng]

Hi, I write a simple python script for visualization at #37. Please have a check about whether it is correct.

Hi, many thanks for your help. I have merged your python script and also updated the matlab code.
It seems this issue can be closed.