D-Lighting Curves and video?

macgregor said:
Keep playing with curves. Sadly you won´t get any extra DR by doing that...
Click to expand...

Not extra DR, but a redistribution of info in the compressed file, such that you get more latitude in grading. In other words, Samuel H's s-log - like image will (theoretically) be more tunable and correctible. Hopefully you'll be able to reshape the flat image into the one you want in post before it starts to fall apart. Actually, given how "thin" the "negative" is, it seems it would still be best to develop a custom picture profile so the recorded image ends up the way you want the final, with information distributed in the right places - or as close as practical.

BUT, if the uncompressed data stream out of the HDMI can be saved (as Pro Res or whatever) in a high bit-rate codec on an external recorder, an s-log profile might yield the "thick negative" needed for more aggressive grading. Hopefully with significantly more latitude than the H.264 compressed file recorded in camera.

Still, it looks like even the uncompressed output is limited to 8-bit. Testing will tell how much optimally shot and recorded images can be pushed around. In the meantime, I just rent whatever I need and leave the experiments to those who have more time and skill than I do. I'd love it if the D800 turns out to be usable as a still and video camera for my work. But until that's proven, a dedicated 10-bit video camera will stay at the top of my "Buy if I decide to own equipment" list.
 
not only that: there's actually more recorded DR in video mode

I can't increase what the sensor sees, but the Nikons usually clip at least 3.5 stops of DR in the shadows in video mode (D7000 has 14 stops of DR in RAW stills mode, zacuto measured 10.3 stops in video mode)
by bringing those clipped shadows back to life before going to the encoder, the recorded DR is increased

it's the same for HDMI-out: if the camera sends out clipped shadows, that DR is dead; if I can bring them up before they're sent to the HDMI recorder, that's extended DR

check these tests and tell me if Flaat_4 is not delivering a lot more DR than any other option (e.g. CineStyle and Marvels Cine): http://www.similaar.com/foto/flaat-picture-styles/real-world-tests-2.html
bringing those stills into your editor and looking at the waveform would be the best way to look at this issue
 
Last edited:
this may be a semantics issue... some people use "dynamic range" to refer to what I would call "codec's color depth" - I'm definitely not altering that (except against CineStyle, because that one actually truncates it)

e.g. Bufo55 said: "Hopefully with significantly more latitude than the H.264 compressed file recorded in camera."
these are two different things: how deep your codec's color space is, and how far apart shadows and highlights can be (in the real world) without any of them being clipped
I can't do anything about the former, I'm definitely expanding the latter (yes: with curves)

btw: expanding the latter without expanding the former means there are fewer steps of brightness for each stop of DR (latter), so banding may appear more easily
Stu Maschwitz advice for this is: gentle denoise, gentle add noise/grain
 
(btw: no idea how much noise it will show in video mode, and I still have to smooth it out quite a bit, but if you've got a D800, get ready for something that feels almost like hdr...)
 
it's nearly done, but not quite, and I have to go to work now
I will release this preliminary version today, with just one profile, based on Portrait, and still not as smooth as it should be
it's not the most exact measure ever, but I've rated it at 13.5 stops (the highlights and shadows are mostly done, the work that's left is in the mild shadows)
 
Samuel H said:
not only that: there's actually more recorded DR in video mode

I can't increase what the sensor sees, but the Nikons usually clip at least 3.5 stops of DR in the shadows in video mode (D7000 has 14 stops of DR in RAW stills mode, zacuto measured 10.3 stops in video mode)
by bringing those clipped shadows back to life before going to the encoder, the recorded DR is increased

it's the same for HDMI-out: if the camera sends out clipped shadows, that DR is dead; if I can bring them up before they're sent to the HDMI recorder, that's extended DR

check these tests and tell me if Flaat_4 is not delivering a lot more DR than any other option (e.g. CineStyle and Marvels Cine): http://www.similaar.com/foto/flaat-picture-styles/real-world-tests-2.html
bringing those stills into your editor and looking at the waveform would be the best way to look at this issue
Click to expand...


Out of all those ones only the technicolor one throws more DR, since it is not just a profile made by someone playing with curves.
I've played before with the nikon profiles on the D7000 a few months ago, to see if I could fix the ugly porsterization of the last stop in dark environments by rasing the black levels and gamma curve. The results were terrible. The same level of posterization would appear so there was no improvement in shadow details.
 
My $.02 is that curves are never going to remove noise from the shadows, but they might be able to add some extra detail to the highlights. That's what knee does in a traditional video camera.

So while curves aren't going to add more DR to the sensor, but they might be able to make how that DR is recorded more useful. For example, this is what Sony's Hypergamma settings do.

That said, I also have to think that 13.5 stops is a pipe dream. Sorry. But I'd be really surprised that Nikon's engineers would allow a signal with that much DR anywhere near 8-bit quantization. It will be way too "steppy." Also, just eyeballing it, I don't see 13.5 stops in any of the footage that's been posted. More like eight or nine (being generous).

Could be wrong. Would actually love it if I were.
 
You'll see: it has drawbacks (noise, banding), but it works
BTW I need 2 people with a D800 to create the easy-to-install file and test it. Send me a pm
 
Just for reference, here are the Milk and TassinFlat + LUT curves. I like the standard Milk preset the best, as the other one is just too flat (yes, it's possible ;) ). The colors also appear nicer, to me at least, than the TassinFlat preset.
 

Attachments

ok, I've got a beta version ready to be tested: http://www.dvxuser.com/V6/showthrea...te-of-picture-controls-for-Nikon-DSLRs-beta-1

I know it can be too flat: what I've made here is a specialty picture control, just for the sake of knowing how far the camera can go; I've rated it at a little over 13 stops (maybe 13.2; I couldn't make it to 13.5 and keep it smooth); if you don't need as much DR as possible for your shot, using this is just silly

if this one works, I'll get working on the workhorse options (it may take a while, though: working with Nikon's viewNX2 software, and without having an actual camera to play with, is slow and painful)
 
I experimented with the custom curve feature in picture control utility on my D700. For testing I created an extreme custom curve that would be immediately recognizable. It seems that the custom curve is not applied in-camera even though it is definitely in the picture control file and works with View/Capture NX2. It just doesn't appear to apply it to JPEG files in the camera. Does the D800 also work (or, not work) this way? For video especially, it would be nice if the custom curve were applied in-camera before conversion to 8-bit video.

I'm interested in the D800 video capabilities and would like to know if picture control custom curves are processed in-camera. It would be great if they were. Unfortunately nobody has been able to answer this question so far on the various forums I've posted to.

If anybody has a D800 and could test this out with video I'd be very grateful. An easy way to test it is to create a custom curve that looks like a sine wave where mid-shadows are raised and mid-highlights are lowered (sort of a sideways S curve). Such a curve makes images look sort of posterized, so they are very easy to see.
 
cbrandin said:
I experimented with the custom curve feature in picture control utility on my D700. For testing I created an extreme custom curve that would be immediately recognizable. It seems that the custom curve is not applied in-camera even though it is definitely in the picture control file and works with View/Capture NX2. It just doesn't appear to apply it to JPEG files in the camera. Does the D800 also work (or, not work) this way? For video especially, it would be nice if the custom curve were applied in-camera before conversion to 8-bit video.

I'm interested in the D800 video capabilities and would like to know if picture control custom curves are processed in-camera. It would be great if they were. Unfortunately nobody has been able to answer this question so far on the various forums I've posted to.

If anybody has a D800 and could test this out with video I'd be very grateful. An easy way to test it is to create a custom curve that looks like a sine wave where mid-shadows are raised and mid-highlights are lowered (sort of a sideways S curve). Such a curve makes images look sort of posterized, so they are very easy to see.
Click to expand...
It's supposed to apply the curves to JPGs, just like Portrait, Landscape, Vivid, etc. They're all the same kind of file. Have you tried in-camera editing, using your custom curve?
 
I figured it out - it turns out that the curve is not applied unless active D lighting is turned off. Hopefully, the curve works in video mode with the D800.
 
the curves should be applied to RAW data to create JPG files, but it seems they are actually applied over a simplified (transformed) version of the original data

this is the test (shot by Latyshev on a D800), maybe somebody else can find a different plausible explanation: http://vimeo.com/39543745

I'll ask Latyshev to shoot a video test with Flaat_13p_beta1 and take a RAW still with the same settings; that way we could see if images are processed differently in-camera and in-viewNX2

edit: it would be great if everything could be solved by switching D-Lighting off...
 
Last edited:
I assume image data would pretty much have to be converted to YUV format before applying curves, saturation adjustments, etc... Hopefully, though, the reduction to 8 bits per channel happens afterwards.
 
Samuel H said:
the curves should be applied to RAW data to create JPG files, but it seems they are actually applied over a simplified (transformed) version of the original data

this is the test (shot by Latyshev on a D800), maybe somebody else can find a different plausible explanation: http://vimeo.com/39543745

I'll ask Latyshev to shoot a video test with Flaat_13p_beta1 and take a RAW still with the same settings; that way we could see if images are processed differently in-camera and in-viewNX2

edit: it would be great if everything could be solved by switching D-Lighting off...
Click to expand...

Yo ya lo dije... lalalralalalaaaa ;)
 
@macgregor: malditos roedores...

in my defense, in theory there's nothing that precludes the successful use of curves to get, in video mode, all the DR that the sensor captures in RAW stills
the only reason this is not working so far is the dumb implementation by Canon (with a mandatory hard clip that leaves you only 11.5 stops of DR to play with; hopefully ML can move this clipping point, I'm playing with that today) and Nikon (with something that seems pretty similar, a bit wider but with very little color depth in the shadows, and with different image processing in-camera and in-PC)

but yes, so far, you're right (and I know you'll agree with me that this is bad news)
 
Samuel, it's not really such a dumb clip (and I'd be surprised if the clip even allows 11.5 stops out of the camera). Trying to compress 13.5 stops, which people seem to be saying the sensor can provide, w/ an 8-bit codec is not a good idea, as it would be susceptible to banding.

Also video displays don't have that kind of dynamic range, which means opening wide the Pandora's Box of display LUTS to get the display to be WYSIWYG.

But thanks a lot for your efforts and you might find a way through a hack like you mentioned. Thanks again!
 
You must log in or register to reply here.
Back
Top