DrDre

Glad to hear you're home, poita! Take it easy, as they say...

Colek said:

DrDre said:

Colek said:

Hello,

I would want to really thank you for this great software  you made here, it will help me a lot in my future projects.

As I presume, I could use it to color correct deleted scenes to match the movie from other source, is that right (the color prediction correction)?

I would need to extract whole movie into .jpg and color correct them frame-by-frame using this, yes? For best results I propably should use new color correction model for each scene?

I tried it quickly with this movie I have been wanting to color correct and work on for a very long time now, but considering the colors it was really tough work for me. Your tool will really help me a lot :)

Source (horrible red tint)

image removed

HDTV Broadcast (IMO best colors of this movie out there, better than DVDs)

image removed

Color corrected

image removed

As you can see it producted a lot of artifacts on that spot there. What should I do about it? Try to match both frames even more to each other?

Thank you very much again :)

 The best thing to do, is to do the color correction shot by shot, although sometimes it works very well for multiple scenes or even a whole movie if the differences in color are very consistent.

The artifacts seem to be caused by some compression artifacts in the source. Increasing the stabilization parameter might reduce these, so try a value of a 1000, and then reduce the parameter if possible. 

 Thanks for your reply, I will try to set stabilization parameter to 1000 when I am at my workstation PC :)

You think I should firstly go with all the filters for the movie (denoisers, Super Resolution etc. using AviSynth for that) or color correction first?

 Color correction first could be useful, because you can remove any remaining artifacts with filters afterwards.

Colek said:

Hello,

I would want to really thank you for this great software  you made here, it will help me a lot in my future projects.

As I presume, I could use it to color correct deleted scenes to match the movie from other source, is that right (the color prediction correction)?

I would need to extract whole movie into .jpg and color correct them frame-by-frame using this, yes? For best results I propably should use new color correction model for each scene?

I tried it quickly with this movie I have been wanting to color correct and work on for a very long time now, but considering the colors it was really tough work for me. Your tool will really help me a lot :)

Source (horrible red tint)

image removed

HDTV Broadcast (IMO best colors of this movie out there, better than DVDs)

image removed

Color corrected

image removed

As you can see it producted a lot of artifacts on that spot there. What should I do about it? Try to match both frames even more to each other?

Thank you very much again :)

 The best thing to do, is to do the color correction shot by shot, although sometimes it works very well for multiple scenes or even a whole movie if the differences in color are very consistent.

The artifacts seem to be caused by some compression artifacts in the source. Increasing the stabilization parameter might reduce these, so try a value of a 1000, and then reduce the parameter if possible. 

pittrek said:

This automated method looks interesting, I love the colours.

Could it be used for fixing faded 8mm film, e.g. something like this ?

Image removed

 In principle yes, but it depends somewhat on, whether the film meets the underlying assumptions of the color correction model. One of those assumptions is that over the run of a reel/film the fading is similar across frames, and another is that over the run of a reel/film, each of the colors magenta, yellow, and cyan, are equally probable to occur. Consequently, the model needs enough shifted frames, to predict an accurate correction.

Here's a comparison to a frame posted by piota a while ago of one of his scans:

Team Negative1 LPP scan:

Poita scan:

Team Negative1 LPP automated color correction:

Here are a few more comparisons between the automated color correction, and the bluray matched to the Tech IB print:

Team Negative1 LPP scan:

Tech IB:

Team Negative1 LPP automated color correction:

Team Negative1 LPP scan:

Tech IB:

Team Negative1 LPP automated color correction:

Team Negative1 LPP scan:

Tech IB:

Team Negative1 LPP automated color correction:

Team Negative1 LPP scan:

Tech IB:

Team Negative1 LPP automated color correction:

Considering how close these predictions are to the Tech IB colors, while not having been matched to the Tech IB, my conclusion is that the original LPP colors were very close to the Tech IB colors. Consequently, I can only conclude that the Tech IB print is very representative of what people saw in the theatre in 1977. 

So, unless the agreement is a huge coincidence, which I think is higly unlikely, I think this is the first solid evidence of what the actual colors were of Star Wars back in 1977, when it was first released. 

So, how do these color corrections compare to the bluray matched to the Tech IB print colors?

Tech IB:

Team Negative1 LPP automated color correction:

Tech IB:

Team Negative1 LPP automated color correction:

The predicted colors are a close match to the Tech IB colors, which includes the slight green shift in the second frame. This seems like too much of a coincidence, so it appears the green shift was also present originally in other prints of Star Wars, and is probably part of the original color grading of the film.

and some preliminary results for the frames piota posted for TESB. Since these are based on only two frames, the predictions are expected to be less accurate.

I've further improved the parameter free color correction, for correcting color shifted prints. Here are some results for Team Negative1's reel 1 preview:

As long as you have enough pixels, and the reds, blues, and greens have to same proportions relative to each other, it should work oke. 

thorr said:

Thanks for reposting 1.2 DrDre.  It's really awesome and fun to use!

Back to my previous question with Bruce Willis and using a different frame that doesn't line up properly, how did you use multiple different parts of the frame to build the color correction model?  With the tool, I could grab a small piece of skin or something white, etc. but I am not sure if that is the right approach.  Thanks!

 You can grab parts of the frame that are very similar, but it won't work if the part is too small or doesn't have enough color information. In principle the method needs to identical frames, with some flexibility, but not too much.

FrankT said:

Is it able to remove the blue tint of Hoth from the DVDs?

 In principle it is meant for correcting old prints, not new color gradings, but it does neutralize the colors if you apply it to a single frame. Normally you would first calibrate the the model on a number of frames of the reel of a faded print. However, here's what it does with Semi-Specialised 2.2:

Here's what the new algorithm predicts for the two frames that were discussed at length a while ago:

AntcuFaalb said:

DrDre said:

I've developed a new method, called parameter free color correction

That's awesome, Dre! How does it work? Can you try it with our 16mm SotS scan?

Also, I'll make sure to try the LUT tonight. I had an unexpectedly-busy weekend.

 It's actually pretty simple. You first go to the appropriate color space for film, namely magenta, yellow, and cyan. You then reverse the color order, take the log10, and shift the average colors, and standard deviation to an appropriate value defined by a set of frames of the shifted print. 

Sure, just post a few frames, and then we'll see what we get.

FrankT said:

What am I looking at? All I see are tinypic signs.

 It was me, I was updating the images.

I've developed a new method, called parameter free color correction, for correcting color shifted prints. It in principle has no parameters, and corrects the color with the push of a button. The only variable is a slight contrast adjustment if it is desired, but it works without contrast enhancement. It's also extremely fast. It's done in less than a second. Here are some results for Team Negative1's reel 1 preview:

Just to make this clear. These colors were reconstructed, without any prior knowledge of what the colors should be. 

Beber said:

Definitely looks better than the 1st clip. Especially in the dark shots. Have you consider using Chewtobacca's Blu-ray based on Wowow broadcast and another HD broadcast to get rid of Wowow subtitle and logo instead of the official Blu-ray? Or maybe both of them, to have the best of the 2, as the official Blu-ray digitally switched on a lightbulb in the end and, if I remember correctly, zoomed in some shots in the submarine hangar. And on top of that, the official Blu-ray has a weird "waving lines" effect, maybe a side effect of optical effects, I don't know, as in the very beginning, on the mountain shot at the bottom of the frame, and on a few occasions later in the movie.

 I did consider it (and tried it), but the bluray color timing, although too orange, is closer to the 35 mm references, it therfore much easier to regrade than the WOWOW. So, since the WOWOW would require a shot by shot color regrade, I decided to stick with the bluray, despite some of the changes. 

FrankT said:

Swazzy said:

Open the test image, then x out of the window that pops up. Open the reference image, and x out of the preview window again. Then building a model will become an option.

 Huh, the instructions didn't mention that. But it works now, so thanks! I'll be back with some of my results.

 I've updated the instructions. 

Here's the new video sample  for the latest color regrade (available till the 12th of October 12), and I think it's a massive improvement over the previous version. There's great detail, and the colors look spot on:

http://we.tl/iOO9eCO4Rn

FrankT said:

I'm getting frustrated - I can't get this tool to operate properly! I've imported the test and reference images, but the "build colour correction model" button doesn't light up. I think I might be doing something wrong. Can you help?

 The tool expects you to correctly crop the test frame, and reference frame. You don't have to, but in that case you have to close the figure window before being able to continue. 

I will make v1.2 available again, because I'm still waiting for some feedback on the LUTs I created.

Link for download: http://we.tl/CKTV8zX6cf (available till the 12th of October) 

thorr said:

I personally think you had it right to begin with at post 12, 13, etc.

 Yes, but although it looks great as a screenshot, there is too much contrast for video, and you loose a lot of detail in dark scenes.

brycebayer said:

DrDre said:

thorr said:

I think you nailed it!  How does it look when applying to other scenes?

Also, can you repost the tool.  I went to download it but it is expired.  Thanks!

 It looks oke, but the colors in the bluray are very inconsistent, so each shot definitely needs separate regrading. 

I'm nearly done with v1.3 of the tool. It should be up by tomorrow, and will allow you to export a LUT for use in programs like Nuke and Resolve. 

 Nice!  How are you making the LUTs?  In MathLab?

 Yes, I wrote a script to make the LUTs in MATLAB. I do have to test if they work properly.

Colors look very good! They're very balanced and seem authentic. 

The last sample looks really great! Thanks for all the wonderful work you've done team!

Here's the comparison against the 35 mm frames, the bluray, and the WOWOW:

35 mm:

Bluray:

WOWOW:

Regrade:

35 mm:

Bluray:

WOWOW:

Regrade:

35 mm:

Bluray:

WOWOW:

Regrade: