GH5 Waveform Display Calibration in VLog

[Tom Roper]

Attached is download link to 6 seconds of VLog-L 2160p30 10 bit 422 straight from the GH5.
Also attached is screen capture of WFM at 03;16. Data level = full. Note the range is 85-940.

Scrub to 05;24 and toggle forward and back one frame. There is a focus shift at that point, You will see the peak level collapse from 925 or so to 768 as it goes out of focus. This implies that yes, 80 IRE is a soft limit, but high pass filtering permits specular highlights above 80 IRE. So I was surprised to see you asking what NR setting because I was about to suggest you set yours to -5 which is in fact what the attached video is set to.

https://drive.google.com/file/d/0B7w...ew?usp=sharing

https://drive.google.com/file/d/0B7w...ew?usp=sharing

[Cary Knoop]

Thanks for posting!
Thinks are certainly not as straight forward as I thought!

You say that high pass filtering permits specular highlights, but what if what we see is either noise or encoding artifacts?

Noise at -5 may suppress valuable noise reduction that cannot be reduced in post. I do not know the internals of the GH5 but I would assume that noise reduction takes place right after the sensor readings and certainly before it goes into the encoder. I would argue that reduction before encoding can be much more efficient than after encoding because it can operate on sensor data and before debayering, assigning a color space and applying chroma subsampling, also encoding efficiency is negatively influenced by the amount of noise in the video.

I attached a clip using the above referenced DCTL to see what is actually above 768 (looking at the RGB code values not the actual luminance).
There is a clear edge and the inside looks like DCT patterns.

I would assume the inside of the lamp is the brightest and not the surface.

[Cary Knoop]

Here is the same clip filtered by 80 IRE actual luminance:

[Tom Roper]

All that could be true. All I was ever saying was that I knew VLog-L was writing code values above 768 (and below 128 as well).

But that small difference between 768 and 940, for HDR (which is all I do), once IDT transformed in ACES to Panasonic V35 and ODT to ST2084-1000 represents a difference in display referred specular brightness of about 400 nits, no small amount. The GH5 is only marginally adequate for HDR anyway, and if the specular highlights are suppressed loses the je ne sais quoi, looks sterile. In HDR, we're only trying to emphasize specular highlights, not blast the viewer with overall display brightness.

So the choice to use NR is yours.

[Cary Knoop]

And you were right about that, it does write above 768!
Question remains for me is that data we should actually use.

It is indeed no small amount.
I certainly would hope the GH5 would have usable values above 768.

Thank you for putting this to the attention of everyone and I am sorry for my harsh words!

[Dan Gunn]

You guys just gave me a headache......just to much info for my few remaining brain cells. Can't I just push the record button?

[Tom Roper]

No problem Cary, it is a good discussion! Now we need for to you to tell us at what level of the NR filtering detail becomes noise.

[Tom Roper]

Originally Posted by Dan Gunn

You guys just gave me a headache......just to much info for my few remaining brain cells. Can't I just push the record button?

If you aren't concerned about punching a hole in the universe, then yes.

[Tom Roper]

Originally Posted by Cary Knoop

Here is the same clip filtered by 80 IRE actual luminance:

This is *REALLY* cool.

[Cary Knoop]

I did some small testing to see how both the noise and the sharpness settings influence the spikes above 80IRE

From largest to smallest spikes:

Sharpness: 0 Noise: -5
Sharpness: 0 Noise: 0
Sharpness: -5 Noise: -5
Sharpness: -5 Noise: 0

Sharpness seems to be a bigger factor than noise.

My current settings are Sharpness -5, Noise 0.

[Tom Roper]

Well that's interesting!

[Roger Keay]

As the narrators in the old westerns used to say " Meanwhile, back at the ranch", I was investigating the V-Log L clip point on the GH5 waveform monitor. I took a simple approach using manual exposure and locking the white balance to 'cloudy'. I only adjusted f-stop, first finding the value that didn't cause clipping on the WFM. I set the camera in record and narrated what I was doing and seeing. I opened up the aperture and described the extent of clipping for each stop.

I followed the recommendations about configuring Resolve and started looking at the clips. Two scenes produced interesting results. I used an ornamental white turtle sitting on my deck as one subject. The turtle occupies about 30% of the width of the screen at the centre. The turtle has coarse details on the shell. I shot from the top looking down. As I increased the aperture, the GH5 WFM clipped on the second dashed line below the main solid top line. On the Resolve WFM, the top of the clip falls right on the 768 line and holds while the background levels move up with a larger aperture. There is a bit of 'fuzz' on the top of the hard clip.

The other scene is shooting upwards through trees leaves and branches, and shows the roof line and chimney of a house. The sky was high white haze. The GH5 WFM showed the same kind of clipping as the turtle shot when I increased the aperture. Now it gets interesting. The beginning of the clip (f13) brushed the 768 line on the Resolve WFM with a fuzzy edge going above the line. At f10, the GH5 WFM showed clipping but Resolve WFM had transients over 832. At f6.3, no sign of clipping on the Resolve scope with transients visible at 896. The main waveform was solidly over 800 where the bright sky was most visible.

I decided to investigate the trees and branches scene more closely. The sky was uniformly bright behind the trees due to the high overcast. I used a power window to isolate a portion of the image where the sky was the primary feature. I used the curve to set areas outside the window to 0. The WFM showed the corresponding area with a distinct flat line at 768. The branches and leaves are creating values above 768 but the sky is clipped. I moved the window around the screen to find peaks. In areas where the leaves and sky occupied similarly sized areas the peaks were visible. I don't know if they are the result of a low pass filter driving the clipper or if they come from artifacts, possibly from edge sharpening. Considering that a major goal of HDR is to preserve speculars, it makes sense that the clipping mechanism would allow finer detail to pass with less attenuation.

A third shot was red flowers. As I opened the aperture, the red channel viewed on Resolve started to clip around 800. The clip looked softer than the sky but the flower petals were not a uniform brightness so a bumpy edge of varying intensity makes sense.

The processing in the GH5 going into the V-Log L profile seems anything but simple. However, if the GH5 WFM is not clipping then the image is probably suitably exposed with bright transients protected.

[Dan Gunn]

Originally Posted by Tom Roper

If you aren't concerned about punching a hole in the universe, then yes.

Hmmm, that might be a real trip!

[Cliff Totten]

There is a chance that im totatly wrong here but I'll toss this out for anybody to confirm or correct:

Its my understanding that VLog is a fixed gamma like Cineon and Sony SLog-3. (Slog-2 is not fixed but individually mapped to each sensor model)

If this true, than VLog itself is not clipping or maxing out at 80 IRE. VLog does not place this limit in any way. (Aside from it's normal gamma compression nature) Its actually THIS particular GH5 image sensor not being able to deliver a significant signal over 80 IRE inside VLog with the exception of stray highlights or noise or whatever.

VLog is simply taking all the dynamic range the GH5 sensor can feed it. Obviously, the VLog bucket or curve still has lots of headroom leftover above 80 IRE "if" the image sensor were able to produce it.

So "maybe", if im not wrong, its really the GH5 sensor that cant output anything significant registering higher than 80. In theory, if the GH5 had a 14 stop sensor than we'd be looking at over 90 IRE in the same exact VLog curve.

I dunno. Does this make sense? All the numbers we are talking about here are not really "VLog" numbers, they are more like "image sensor" output numbers.

[Roger Keay]

I think you must separate V-Log encoding from camera video pre-processing. I think the requirements for a consumer camera are significantly different than a professional camera.

The GH5 is a product of Panasonic's consumer division, not the professional camera group. The camera is meant to be sold to hundreds of thousands of customers, many of whom will have a casual interest in videography and little technical knowledge. The GH5 must be consumer friendly and play nicely with regular consumer televisions.

High dynamic range television sets are just beginning to establish their place in the market. Consumer sets are only capable of limited HDR performance because the power supply in the set limits the total area of the display capable of achieving high brightness at any given instant. In practical terms, the set can deliver full screen area brightness up to the usual level for SDR (100 nits) plus limited area highlights (100 - 1000 nits). These limitations are not a severe problem as most of the highlights come from specular reflections or point sources at night. HDR can deliver a night street scene with good detail in the darker areas while traffic lights and neon signs are rendered in bright, saturated colors.

If a company wants to deliver an HDR consumer camera that works well with consumer HDR displays then it must deal with the issue of highlight area versus brightness. The simplest approach might be to use a scaled variation of the camera's normal SDR processing to keep large areas in check and separately high pass filter the video to create the HDR details that can be added to the SDR portion of the image. The resulting HDR signal can be encoded into HLG or V-Log L formats.

A professional camera designer would have no reason to take display limitations into consideration for high V-Log levels. Recorded images are going to end up in the hands of a colorist who will adjust them to match the capabilities of the target display device. Only one kind of V-Log decoder is needed in an edit program because the code values have the same meaning in V-Log or V-Log L.

I have no knowledge of the video processing that Panasonic is using with the GH5 so please understand this is all theoretical. I am simply trying to match my observations of GH5 V-Log L signal behavior from with a possible explanation.