GH4 YAVL - Yet Another VLOG-L LUT

[joe12south]

My love/hate relationship with V-LOG L continues.

There's no doubt that the extra latitude V-LOG L affords can be very useful, but the way it compounds 8-bit H264 encoding errors makes it very fragile when recorded internally. After a lot of testing, and a lot of tweaking, I think I've developed a decent input LUT that represents color more accurately than Panasonic's 709 LUT, and takes some steps to mitigate noise and encoding errors (no LUT can eliminate them, the best you can hope for is to make them less noticeable.)

I've still got more work to do (dialing in magenta through yellow on the GH4 is ridiculously hard...as we all know from struggling with its unpleasant skin tones) but I'm getting close. Take a peek and let me know what you think.

---

First, we have my personal Cine-D settings designed to hold as much latitude as possible. Note that our "subject", the chart, is exposed to protect it's highlights. This yields a slightly underexposed image in which the areas of direct sunlight are still completely blown out (unrecoverable.)

---

Here's the V-LOG L version, exact same exposure settings, with the VAVL LUT applied. Note how there is an extra stop of highlight information, and how the colors are more accurate. Many subtle changes to both hue and saturation throughout the spectrum. On the downside, note the additional noise and the green/magenta macroblocking on the back wall.

---

Now, let's correct the exposure and increase saturation to a natural level. Note how we are able to do so without causing the problem areas to jump out too badly.

 

[stoneinapond]

Excellent taste in beer…...:happy:

Looks interesting. Personally I'd like to see the mids lowered a bit, but the detail in the back window is encouraging.

Thanks for sharing.

 

[visceralpsyche]

Nice work I've emailed you v154 BETA of my own V-Log LUT for comparison. I'm curious to see what you think!

Cheers,

Paul

 

[joe12south]

Looks interesting. Personally I'd like to see the mids lowered a bit, but the detail in the back window is encouraging.

Unless I screwed-up, the mids should be spot on in terms of luminance. (I.e. 50% grey is right around 128.) Remember, this is an input LUT, so the goal is to be as accurate as possible. After applying this LUT you can then use any REC709 LUT, or do other grading, to taste.

 

[stoneinapond]

Remember, this is an input LUT, so the goal is to be as accurate as possible.

Thanks for the clarification. In that case I'd say it's looking good. :beer:

 

[joe12south]

Here's a comparison between CIne-D, version 154 of Paul's LUT, and YAVL.

- Cine-D handles muted colors, like the back wall better than V-LOG L. At the same saturation, V-LOG L has more hue errors (no surprise.) YAVL takes steps to minimize it, but it's impossible to remove completely.
- Panasonic's color science is significantly green/magenta deficient. YAVL goes further than Leeming to correct this. (Many people find "accurate" green to look too saturated, so Leeming my be more palatable on its own.)
- Both Leeming and YAVL better hit the skin tone patch than Cine-D.
- YAVL looks to be more aggressive at desaturating shadow noise.

 

[Greg_E]

Even on my crap monitor, look at the noise in the gray cube in the last images.

 

[joe12south]

Even on my crap monitor, look at the noise in the gray cube in the last images.

What specifically are you pointing out? I don't see the V-LOG L noise as appreciably worse than the Cine-D sample. Underexposed, the GH4 is always quite noisy in the shadows, even at native ISO 400 (the ISO for these shots.)

 

[Greg_E]

Just noting that there is visibly more noise in the other two images. The edges on the cine sample are also much sharper, might be nr and sharpening in the cine setting, might be just a focus difference between the two recordings. Either way, you asked for comments on this comparison, I have therefore commented on what I see. To me, this extra noise could be a real problem going forward, to others maybe not so much. Either way, it should be noted so that the end user can decide for themselves.

 

[wwjd]

probably discussed ad nauseum, but couldn't the blow out window in cined be reecovered using "Highlights" reduction, or lowered a 1/3, and or raising the Pedestal to make the gray match? Is there really NO WAY to recover that blown out detail than using VLOG?

Also, what color is your back wall in reality? The VLOGed images seem to have more greenish tint overall to me - especially those back walls. (monitor calibrated with one of those EYE thingies, SEEMS pretty accurate)

 

[joe12south]

Just noting that there is visibly more noise in the other two images. The edges on the cine sample are also much sharper, might be nr and sharpening in the cine setting, might be just a focus difference between the two recordings. Either way, you asked for comments on this comparison, I have therefore commented on what I see. To me, this extra noise could be a real problem going forward, to others maybe not so much. Either way, it should be noted so that the end user can decide for themselves.

Thanks, Yeah there is a difference in the sharpening settings between the two (Cine-D was at -3, V-LOG L was at -5).

 

[joe12south]

probably discussed ad nauseum, but couldn't the blow out window in cined be reecovered using "Highlights" reduction, or lowered a 1/3, and or raising the Pedestal to make the gray match? Is there really NO WAY to recover that blown out detail than using VLOG?

Unfortunately if you under expose the Cine-D two stops (which is what you'd have to do to get that much highlight information) the overall image will be significant crushed such that when you recover, you're left with a noisy, muddy mess. Even if you significantly lift the shadows and pedestal in camera, it doesn't compare to a LOG curve. Additionally, the colors really go haywire.

Also, what color is your back wall in reality? The VLOGed images seem to have more greenish tint overall to me - especially those back walls. (monitor calibrated with one of those EYE thingies, SEEMS pretty accurate)

The back wall is a very pale, fairly neutral beige. The Cine-D footage captures it most faithfully. It's going green in V-LOG L because of the YUV encoding errors that get exaggerated when expanded to REC709. I've yet to find a way to completely neutralize that without negatively impact "memory" colors.

 

[wwjd]

Well, I think what you are doing looks great. Vlog hasn't worked for me yet with the crunch, banding, and greenism. Are you making this for fun, or free, or profit? I haven't messed with Paul's LUT either, but you guys are WAY APPRECIATED doing this sort of stuff - where Panasonic has not

 

[joe12south]

Well, I think what you are doing looks great. Vlog hasn't worked for me yet with the crunch, banding, and greenism. Are you making this for fun, or free, or profit? I haven't messed with Paul's LUT either, but you guys are WAY APPRECIATED doing this sort of stuff - where Panasonic has not

Thanks. I'm personally still going to avoid V-LOG L unless I'm using an external recorder, but I want to have this LUT in my toolbox in case I need to record it internally. (Some scenario in which I can't control the lighting and simply must have the extended dynamic range.)

I'm not happy with the results yet, but I wanted to share at this stage to see if I'm on the right track. I need to upgrade to a more sophisticated (and seriously expensive) LUT creation tool to do the final manipulations. If I have success, I'll make it available somehow.

An even bigger, and I think more useful, project that I am currently undertaking is a LUT and noise profile combo that will recover under-exposed ISO1600 tungsten footage. If I can pull that off, that will *really* be something. Initial tests are promising.

 

[joe12south]

This is very early, and still needs a ton of work, but here's a sneak peak at CineDlite Recovery:

Unprocessed file

LUT applied

Finish applied

 

[Greg_E]

Oh, that is a huge improvement! I bet the final would like just fine with moving images.

 

[joe12south]

Oh, that is a huge improvement! I bet the final would like just fine with moving images.

Thanks, it's usable, but a lot of the colors (especially the reds) are wrong. If you had to intercut this with correctly shot footage, it would be really hard to make it match. Pretty much impossible in any grading tool I know of, including Resolve. But, using a LUT editing tool to directly manipulate very discrete sections of color, I think I can get this as close as is possible given the limited data that is recoverable. I'll definitely let everyone know if I have success.

 

[wwjd]

what was your "Finish" applied? De-noiser?