Face Refinement doesn't work on profile or 3/4 faces/ JOG

While I love the new face refinement tool, I have found (and also seen in the linked below youtube tutorial). With profile faces and 3/4 static faces the mask extends too far, as if it's anticipating the shape of the face to be looking more directly towards camera.

Perhaps an easy fix would be to include a rotate function to the mask, or allow a warp of part of the mask.

Also for head turns, key framing this rotate or warp could fix this situation.

Maybe this concept needs to thought of as a starting point, and not an automatic filter

(Except in the case of interview talking heads)

Allowing some method of removing mask key frames would be needed as the below demo

Shows a few 1 frame jumps in the mask during tracking, that could be smoothed by eliminating the

Frame for that mask. https://m.youtube.com/watch?v=jDPDpWLSNkY

On a side note, please allow the Advanced Panels jog/ shuttle to work in the preview window for editing.

I would love to jog/shuttle to a frame, hit in and out points and insert/ overwright all from my advanced Panel. The new engine makes jog shuttle work so much better - though it seems each jog is 2 frames, not one. Please make this work to leverage our investment in the big panels.

Going to play a bit of devils' advocate here...

The only exposure I have had to date with "14Beta" is what I saw at the Las Vegas Convention Centre. The major appeal to me: Fairlight, which still has some distance to go before release, IMO. There are a couple of featured implementations; secondary "compression" and the sequence/comparison display that is part of, but not exclusive to, the collaborative workflow that are very interesting.

Yes, the face-recognition algorithm works with full-on two eyes, nose, mouth. That's what it does. Its why I have Boris Beauty Studio, which connects to the Mocha planar tracker through its Pixel Chooser. No, it doesn't have eye-bag or cornea-replacement surgery... I believe the Resolve refinement will improve - that is just the team's track record. Its an organic thing we do, and really underlines the difference between Photoshop-ing a still and cinematography.

jPo, CSI

Agree with Joe, think it is a very early Beta especially with the Face Refinement feature. Lots of potential for anything below a L'Oreal beauty commercial

Yes, the face-recognition algorithm works with full-on two eyes, nose, mouth. That's what it does. Its why I have Boris Beauty Studio, which connects to the Mocha planar tracker through its Pixel Chooser. No, it doesn't have eye-bag or cornea-replacement surgery... I believe the Resolve refinement will improve - that is just the team's track record. Its an organic thing we do, and really underlines the difference between Photoshop-ing a still and cinematography.

I agree totally with Joe. I think ultimately this is a "gee whiz" feature that may not be that practical or useful in real life. I also think it would be possible to create a PowerGrade that essentially does the same thing (but would have to be tracked separately).

Tracking in 3-dimensional space is very, very complex and inevitably involves some manual adjustment. There is no way to do this as a "set and forget" operation.

Boris Beauty is a very interesting plug-in, though I've only used it a few times. I think giving people options in addition to -Midtone Detail, Mist/Blur, and SNR softening is a good thing. Real beauty color correction requires a lot of thought and experience -- the guys who do this kind of commercial work are among the best in the world.

Thanks for all the thoughts on it. My point in posting was to provide some feedback to the team and ideas for making this a viable tool for other than interviews, etc. Warren's thought was a great one, that with improvement this could be used for many projects other than high end beauty , etc.

I'd also like to see the individual zones, take an input filter, so for example I could use my beauty box
Or some other skin/detail smoothing plugin inside FR's cheeks, forehead individually.

I'd also like to see the individual zones, take an input filter, so for example I could use my beauty box Or some other skin/detail smoothing plugin inside FR's cheeks, forehead individually.

You can do this now by creating individual masks, even within one node.

I've tried masking a face on a node then running face refinement but it still ends up jumping between faces which clearly doesn't work!

The Face Refinement tool is an amazing tool when it works. On 4K x-t2 footage some clips track well; others not so much. The effects seem to be happening several degrees below the mask (eye, chin, cheek) and not on the actual keyed out mask.
Strange behaviour. This tool though. When it works; will save time.

This tool though. When it works; will save time.

That's very true. I had a feature about 6 weeks ago where for about 20 shots, we needed more eye illumination and I was able to just use that and a tiny (and I mean tiny) bit of sharpening, and it made all the difference in the world.

Be warned it's very, very easy to overuse this stuff.

Be warned it's very, very easy to overuse this stuff.

I totally agree, and within good experiences which works great, I think it's the main danger actually.

I happened to use it successfully on a dozen of clips this week. (If it had not been faces, I would have grouped them for a post-group grade btw). It was for a tiny bit of blush and cheeks stuff... tiny...
Then I tried a comparison between this node and another one with skin selection, PW track of cheeks etc and correction.
At the end of the day, Face Refinement can do the same if the shot is easy (not too much head turns, sudden moves etc...), and even quite good.
But : 1. In case of well prepared shots (turns, movements, perspective/face cam etc...) and 2. same (user tentation) problem as you may face with NR : it's very tempting to go beyond what's really needed.

If you take your face retouching seriously.. then all these Swiss Army Knife solutions are eventually (more or less) the same and just not good enough in complex scenes or tough situations.

A talented colorist named Benoit Cote from the Lift Gamma Gain forum published a workflow + node tree explaining his take on Frequency Separation using Resolve. he modified it since and I believe that thread in LGG is still very alive.

Its not a simple technique, but mastering it brings the best (and most natural) results I've seen.

Worth learning it too.

Node tree looks somewhat like this (though non active in this picture).

Screen Shot 2017-10-20 at 13.17.50.png (730.78 KiB) Viewed 5708 times

A talented colorist named Benoit Cote from the Lift Gamma Gain forum published a workflow + node tree explaining his take on Frequency Separation using Resolve. he modified it since and I believe that thread in LGG is still very alive.

Benoit is very good, but actually ResolveFX Sharpen is a frequency-splitting sharpening tool that BMD gives away for free in Resolve 14.

"Magic Sharp" with BorisFX does the exact same thing. The trick is that it can increase apparent sharpness without affecting noise, which is a fairly clever achievement. Normal enhancement can't distinguish it unless you qualify the high frequencies, mids, and lows separately. (Users in Resolve 12.5 would need to use Magic Sharp or Benoit's multi-frequency technique.)

I much prefer using FX Sharpen or Magic Sharp because it takes about 1/10th of the time to set up and is done entirely in one node. Sharpen Edges and Soften & Sharpen are more specialized tools for certain kinds of situations. There's lots and lots of adjustments that can be done, but the factory defaults actually are pretty close much of the time for me.

Benoit is very good, but actually ResolveFX Sharpen is a frequency-splitting sharpening tool that BMD gives away for free in Resolve 14.

"Magic Sharp" with BorisFX does the exact same thing. The trick is that it can increase apparent sharpness without affecting noise, which is a fairly clever achievement. Normal enhancement can't distinguish it unless you qualify the high frequencies, mids, and lows separately. (Users in Resolve 12.5 would need to use Magic Sharp.)

I much prefer using FX Sharpen or Magic Sharp because it takes about 1/10th of the time to set up and is done entirely in one node. Sharpen Edges and Soften & Sharpen are more specialized tools for certain kinds of situation.

Thanks Marc. I'll check both.

I was pretty skeptical about this new plugin, but I have to admit that it's saved my butt a couple times since I gave it a shot. Admittedly, it's only good, and will only function properly, in specific situations. It really does need your subject to be facing the camera throughout the shot. But, when that criteria is met, at least in my experience, it works really well. I just finished a series of corporate vids that featured a presenter speaking to cam. One angle was shot with a Red Epic-W that got us some really nice fallof on the face, and the other a Sony 8 bit cam that almost drew a line down the middle of the face. The difference in falloff between these 2 angles/files was significant and this tool really helped me with contrast between light and dark sides of the face, controlling highlights, and adjusting color. So, I'm a fan and can hardly wait til it's improved and becomes more robust.