Color matching and prediction: color correction tool v1.3 released! — Page 43

Hal 9000 said:

I’m trying to use this for two different sources for ESB, one from 2004 and another the 19SE. I align the images as best I can before generating the two frames to feed into the tool, but I can’t get it to be exact.

Would it be better to use a higher smoothing parameter? I haven’t seen DrDre recommend anything higher than 0.1, but it can go as high as 1. If the two images are not totally identical (same frame, but the sources aren’t pixel-perfect), is there a drawback to using 1 as the smoothing parameter?

A higher smoothing factor is less exact, but reduces the incidence of artifacts. I generally rarely use a value over 0.2. For more exact matching use more color spaces and a smoothing factor of 0.001. The frames have to be virtually identical to prevent artifacting.

Hal 9000 said:

How would you correlate the number of color spaces with the exactness of the image alignment, and same question for smoothing parameter?

More color spaces lead to a more exact match. A lower smoothing parameter does the same.

Sounds like a higher smoothing parameter is better when the sources aren’t pixel-perfect. Is a higher number of color spaces always preferable, with the only downside being time?

Yes, this is true.

Also, would it make any sense at all to create a collage of multiple frames (twice, per source) and feed that into the tool for a more informed LUT to export?

Yes, but use plenty of color spaces (>20) and set the smoothing factor to about 0.1.

Hmm. Tried installing the MAC version from the link Dr Dre provided above (thank you again!) but the installer won’t even run on my Big Sur iMac and gives me an error when installing on my High Sierra MBP. It tells me “…Contents/Resources/Bundle.zip does not exist”. Any advice from Mac users?

Hal 9000 said:

I’ve just been running it within a Windows VM.

Thank you for the tip! Any particular VM suggestions?

Hal 9000 said:

I’ve used VMware for years and had no problems. VirtualBox is a good free one I had used in the past.

Also, stitching multiple frames together seems to work really well. However, mammoth images seem to choke it and it does not progress past a certain point. But even with a low amount of color spaces (since a large amount choked it) the results are pretty darn good across a whole movie with very different locations.

Yes, there’s a limit to the size of the image it can handle ~8K. So, make sure to resize the frames to something like 360p before putting them together. This should not affect the color matching result.

Hal 9000 said:

Thank you, that helps me plan how to wrangle them into usable input. Should it matter whether they are stacked horizontally or vertically? It’d be easier and more precise for me to do that than to arrange them into a grid.

No, it doesn’ even matter if the source and reference frames are in the same order. What matters is that the source and reference contain the same color information.

Hal 9000 said:

It… doesn’t matter if the images are aligned? I assumed the way it worked was to compare each pixel with the pixel at the same address in the other image.

If it doesn’t matter what order the images are in, I guess it shouldn’t matter if the two images are even perfectly aligned. If one were slightly stretched or squished relative to the other… that may not matter?

They should be perfectly aligned, but stretched or squished doesn’t matter, because it does not affect the information in the image. A squished or stretched number four is still a four. A number four that is missing it’s top is not a four. Because the method works with color distributions across the image, the precise location of the colors doesn’t matter, and so the order of the processed images doesn’t matter, but they should have the same content.

Hal 9000 said:

I’ll do that. I was erring on the side of more smoothing only because I assumed it mattered how closely the test and reference images were identically aligned.

They should be cropped the same.

If I understand correctly, I could shuffle the pixels of both images into a random order and get the same result.

Exactly this.

That guy with no name said:

DrDre said:

Hal 9000 said:

I’ll do that. I was erring on the side of more smoothing only because I assumed it mattered how closely the test and reference images were identically aligned.

They should be cropped the same.

If I understand correctly, I could shuffle the pixels of both images into a random order and get the same result.

Exactly this.

Why’ll you’re here Dr, why do sometimes weird colored pixels show up in the highlighted areas and never seem to go away on any settings?

It’s an artifact. If you find an example, please let me know. Usually it can be fixed.

I have a couple questions:

1. If you were to match an HDR source to the color grade of an SDR source, and the intended final product will be rendered in SDR/Rec 709, would it be best to an HDR->SDR conversion before using the color matching tool?

2. When matching a source with little or no film grain to the color grade of a highly grainy source, should you apply noise reduction to the latter source before using the tool, or would that not affect the results either way?

TiredEditingGirl said:

I have a couple questions:

1. If you were to match an HDR source to the color grade of an SDR source, and the intended final product will be rendered in SDR/Rec 709, would it be best to an HDR->SDR conversion before using the color matching tool?

It would be best to first convert the HDR to SDR and then do a color match.

2. When matching a source with little or no film grain to the color grade of a highly grainy source, should you apply noise reduction to the latter source before using the tool, or would that not affect the results either way?

The grain will not affect the results, unless it is chroma noise, but even there the effect will be quite small.

Since I had a W10 update, the program no longer wants to launch 😕

That’s annoying. I have W11 and it still works fine.