Part 2 of LS300 review is now up

[Steve Mullen]

Part 2 of my LS300 review is now up at:

https://www.thebroadcastbridge.com/c...h-j-log-part-2

Next to come is grading log in Resolve, Premiere Pro CC, and FCPX.

[Lee Powell]

Thanks for the work you put into compiling this in-depth report on shooting with the LS300's J-Log1 profile. The documentation you provided is very helpful in evaluating the dynamic range of the camera. I think much of it, however, has been presented by JVC in a misleading technical manner that makes comparisons between ITU709 and J-Log1 modes unnecessarily difficult and confusing, in support of a conclusion I find highly questionable: "J-log1 increases dynamic range by 3.2-stops (from 400% to 800%)." This claim is backed up by a chart that I think requires careful examination and deconstruction:



There are several aspects of the scaling used in this chart that in my view paint a misleading comparison between ITU709 and J-Log1 modes as implemented in the LS300. The chart's vertical encoding level scale is calibrated in 10-bit values. As the LS300 is an 8-bit camera, for this discussion I'll divide these values by 4 to use the more familiar 0-255 level scale. The horizontal scale is divided in major divisions of 2-stops, with an arbitrarily chosen "stop-0" reference positioned 4-stops above the level marked "100%" exposure.

1. On the vertical scale, the chart indicates J-log1 max level as 255 and ITU709 max level as about 220. This is definitely not how ITU709 mode records video levels on the LS300. When set at 108%, max level in ITU709 video is 255, and highlights in the 220-255 range are fully recoverable. What's misleading about the chart is that this discrepancy makes it erroneously appear that this top-end "400-800%" exposure range does not exist in ITU709 mode.

2. The horizontal scale is marked "100%" at a point near level 32 in both curves. (There's a bit higher pedestal in the J-log1 curve.) Right off the bat, the first question to ask is: 100% of what, exactly, does this level represent? In my view, this looks like the bottom end of the useful dynamic range of the sensor. The chart shows there are only about 8 recordable encoding levels below this point, spread over a 2-stop range. That's less than 2-bits of color depth per stop, i.e. practically useless.

3. So if "100%" is near 0% black, what do the 400% max DR claimed for BT.709 and 800% max DR for J-Log1 actually mean? Again, the chart paints a misleading picture, implying there are 3.2-stops of additional highlight range in J-Log1 mode. In reality, there are no missing 3.2-stops of superwhite highlights above the max level of ITU709, it reaches the same 100% white level at 255 as J-Log1. The only apparent difference is that the in-camera spot meter reads "400%" at this level in ITU709 mode, and "800%" in J-log1 mode. But wait, there's one more difference - base ISO is 400 in ITU709 mode, while base ISO is 800 in J-log1 mode.

So there's the smoking gun: The numbers in J-log1 mode are scaled twice as high as ITU709 numbers.

To construct a useful comparison chart, the top right end of the ITU709 curve should be shifted up and over to the right to coincide with the top right end of the J-log1 curve, and the horizontal scaling adjusted to the correct mid and black levels for each curve. We would then be able to compare their relative flatness and dynamic ranges using a single consistent scale for both modes.

In practice, however, I've found that JVC's in-camera spot meter is a bit optimistic about recoverable highlight levels. In ITU709 mode, I find that highlights above 360% already reach level 255, and everything above that clips. Likewise in J-log1 mode, highlights clip above 720%. As for what this actually means in terms of useful dynamic range, the practical extent of a camera's DR is not determined by these arbitrary percentages anyway. What counts is how deep into the shadow range the sensor can record before details are obliterated by noise and/or bottomed-out color depth. Both ITU709 and J-log1 curves are so close to horizontal in this region that it's hard to pick a practical cut-off point.

I'd like to emphasize that these criticisms apply to the JVC documents reprinted in your review, not to your highly informative discussion of exposure techniques on the LS300. Moreover, I may have misinterpreted JVC's intended use of its mysterious "800%" J-log1 rating, and would appreciate any clarification you may be able to add.

[Steve Mullen]

Thank you!

I almost left this section out because I could not figure out HOW JVC got their 400 to 800.

My notes ask, what is 400%. If it was peak 709 then what was the dynamic range of 709. Almost everyone said 5-6 stops. Which they based upon five 20% increments -- 0% to 100%. For non broadcast 100 to 110 would add a 1/2 stop more. So 5.5 seemed very reasonable.

A Knee it is claimed can add 1.5 to 2 stops. So my 8+ calculation also seems reasonable.

IF this was 400% -- then I kept trying to understand 400% "greater than what." Likewise, what does Sony mean by 1300%. Looking for information on Sony, I stumbled upon the scheme I used in the chart.

The only thing supporting its validity is that I was sure of the 11+ maximum stops. And where 400% fell was at 8+ stops which matched my earlier 8+.

I'll read your post about the digital numbers again. I took the inclusion of 10 bit values as an easy way to use Resolve for grading. And/or it is the SDI output!

[Steve Mullen]

Originally Posted by Lee Powell

But wait, there's one more difference - base ISO is 400 in ITU709 mode, while base ISO is 800 in J-log1 mode.

So there's the smoking gun: The numbers in J-log1 mode are scaled twice as high as ITU709 numbers.

I had read someplace that the ISO could be as low as 200. So I kept checking to see if there was any difference in ISO between 709 and JLOG. The readout was always 400.

If you are basing the 400/800 difference based on what other cameras do -- I too wondered about this. There seem to be 2 schemes. Actually change the possible ISO values so the user can see it force a higher ISO or increase the gain by 6dB without the user knowing.

<< In either case, noise might increase, but running with a 6dB gain on any camcorder I've used never shows an increase in noise at +6. Not until +9 or +12. So it didn't worry me. (I really don't care about noise or blacks because in the 30 years of video cameras everything looks so clean to me. Not to mention starting with Kodachrome ASA 25.) I care, especially in HD and UHD about blown highlights. So even the promise of greater DR is a winner for me. :) >>


I see what you mean that the digital values could be doubled -- which would account for 1 stop higher JLOG readings. That seems like a very stupid error.

Unless, of course, JVC does not change the ISO. This is what I think is true. Which is why I included the comment to try 800 if the Histogram won't go fully right. Since I never use the Histogram for setting exposure -- it didn't make any difference to me. I looked only for a nice looking Histogram -- and because it blocks the VU meters, I turned it off.

So I'm not sure I agree that the extra stop is false.

[Lee Powell]

I noticed the ISO change from 400 to 800 when switching from ITU709 to J-log1 and assumed that meant 800 was the minimum ISO in J-log1 mode. It's actually 400, the same as ITU709. Seems there's an undocumented mechanism in the firmware that alters your preset L-M-H Gain settings when switching back and forth between ITU709 and J-log1 modes.

If L Gain is set to ISO 400 in J-log1 mode, it will remain unaltered when you switch to ITU709 mode, but the preset ISO values stored in the M and H Gain settings will be cut in half. If you then revise the M and H Gain settings in ITU709 mode, they will be doubled when you switch back to J-log1 mode. If you then change the L Gain to ISO 800 in J-log1 mode, it will be cut in half to ISO 400 when you switch back to ITU709 mode. Thus the firmware actively conspires to double your current ISO setting when switching into J-log1 mode, though you can override this for the special case of ISO 400 in J-log1 mode.

At this point, I'm not sure whether this ISO doubling also plays a role in the 400-800% switch in max spot meter readings as well.

[Steve Mullen]

"Seems there's an undocumented mechanism in the firmware that alters your preset L-M-H Gain settings when switching back and forth between ITU709 and J-log1 modes."

I missed this because after a shooting a gray/white card in 709 I switched to JLOG and likely automatically set 400. Since I was shooting outdoors in Las Vegas I only needed 400 and still needed 1/16th ND.

In 709 increasing gain by 6dB -- on the same gray/white card -- should increase double the values. But, since, for example, meter values (meter set to 800 rather than 400) would indicate a one stop lower exposure -- I don't see how/why the digital values would double.

For an appropriate exposure, the ISO will not and should not change the recorded value. Nor should noise visibly increase.

The difference will come in low light. If ISO 400 will not not get a chart to read the appropriate level (too low) switching from 400 to 800 can enable you to get to the desired exposure. In the former case the digital values will be half what's needed, while in the latter case they will be correct.

[Steve Mullen]

Digital data in 709 and JLOG will always lie in the range from 64 (0%) to 940 (100%) or 1023 (108%). The ls300, according to jvc, records from 96 to 1017. That's how it comes into Resolve.

The key is not a difference in the digital range, it is HOW light levels are mapped to digital values. LOG makes each stop in light more equal in the digital range to other stops. See:

https://www.thebroadcastbridge.com/c...into-log-gamma

Bottom-line I think the chart based on info from JVC and Sony is correct.

A possibly more serious issue is how well 8-bits can carry log data. I think this worry is eliminated by the ls300 only increasing DR by 3 stops over 709 rather than 6 to 7 stops available from S-log2 and S-log3.

My samples from a Sony F7 and Canon C300 II are far flatter and take more time to grade. The jvc is really simple because JLOG is not that different than 709.

[Lee Powell]

Originally Posted by Steve Mullen

Bottom-line I think the chart based on info from JVC and Sony is correct.

A possibly more serious issue is how well 8-bits can carry log data. I think this worry is eliminated by the ls300 only increasing DR by 3 stops over 709 rather than 6 to 7 stops available from S-log2 and S-log3..

While the JVC and Sony charts may be technically "correct" (from a certain point of view), the impression they create is misleading and the claims of increased dynamic range are grossly inflated. The chart below from Sony demonstrates yet another way to misrepresent the data:



Here the vertical scale is labeled in IRE percent while the horizontal scale is ruled in mysterious "100%-of-what" divisions. The evident intent is to show how S-log2 has "1300%" DR while S-log2 has only "1000%".

In this case, the sleight-of-hand is easy to spot. In S-Log1, a light level of 1000% is encoded at IRE 109 while in S-Log2 a light level of 1300% is encoded at the same IRE 109. Now what kind of photographic mechanism could accomplish this sort of mathematical scaling effect, perhaps different values of ISO? Yes, it really is that simple of a trick. Another clue is that the supposedly normalized input level of 100% is encoded at different IRE levels for each curve, a tip-off that the curves are not scaled to the same reference level. Once again, to produce a valid comparison between the curves, the upper right end of each curve should coincide at the same point, both horizontally and vertically.

What both JVC and Sony charts have in common is that they're trying to convince us that the curve with bigger numbers has multiple stops of increased DR in the highlight region. But that's obviously bullshit since any tone curve that's exposed at the correct 100% white ETTR level will record all highlight levels below that point as well. In other words:

Dynamic Range starts at 100% white and extends downward to the point where image details are obliterated in the shadows.

If J-log1 truly has "3-stops more DR" than ITU709, it should be able to record 3-stops of additional shadow detail, just above 0% black. I frankly haven't seen anything close to that in my tests of the LS300, I'd estimate it's not even 1-stop better. And that's not because the shadows are too noisy, it's due to both curves encoding the darkest shadow details with only 2-bits of color depth.

Where I do see a difference in highlights between the two LS300 gamma curves is in the compressed highlight knee of ITU709 versus the uncompressed highlights of J-log1. That, however, does not constitute an increase in highlight DR, it's just a difference in 8-bit encoding that requires different types of LUTs to properly grade in post. If you don't correct ITU709 footage with an inverse highlight knee, highlights will look less detailed than in J-log1 footage. In practice, you don't want to get too close to the limit with either curve, as the top 10% of the advertised highlight range is indistinguishable from white.

[Steve Mullen]

I think you are missing the point that no matter 709 or log, all values from 16 to 255 are used. So of course the peak values are 108%. And black is 0%.

Noise is a function of sensor and processing. Since both curves are so close in the darkest stops I would expect they would show the same noise levels with the JVC. May not be true of other log curves.

709 is a CORRECTION -- to match CRT response curve. Use Excel to take linear-light data and convert it to 709. Note where the stops fall. Then multiply the data by the CRT curve and one is back to linear-light.

Log is not a CORRECTION. It's purpose is to convey the linear-light data from the sensor. Again use Excel to Excel to take linear-light data and convert it to log. Note where the stops fall.

If you are claiming log is a hoax there really is nothing to be said. If you simply pointing out that with JVC sensor JLOG only provides a modest increase in DR over 709 -- that is factually correct. Put in a sensor with 15 rather than 11 stops DR, and different log curves will dramatically increase DR.

[Lee Powell]

Originally Posted by Steve Mullen

If you are claiming log is a hoax there really is nothing to be said. If you simply pointing out that with JVC sensor JLOG only provides a modest increase in DR over 709 -- that is factually correct.

Actually, I pointed out in my first reply that contrary to JVC's ITU709 chart, the LS300 records ITU709 highlights all the way up to encoding level 255 in 108% mode. And no, I'm not disputing the value of log tone curves - with proper grading they can improve both highlight and shadow detail over ITU709.

What I consider misleading are Sony and JVC's claims of several stops of increased highlight detail, which we both seem to agree are at best modest. Moreover, I suspect ISO manipulation is used behind the scenes to make log curves appear to provide an extended highlight range. And I'm still mystified over the "800% of what?" issue.

[Steve Mullen]

"And I'm still mystified over the "800% of what?" issue."

Finally figured out that there is no need of answer

Think of a battery comparison in the form of a horizontal bar chart.

For both batteries the full (1.5V) and empty (0) values are the same. All we want to show is the relative difference in life.

The bars SHOW us that. 2X longer life -- the bar is 2X longer. The X axis is representational. There is no need for values in the chart.

This is the same for our log chart. You see the difference. Since it's all linear there is no marketing BS going on.

But why put in the %? In the camera there needs to be some way of referring to the peak values. A sad face and happy face would be fine, but a bit silly. So % was chosen. A very bad idea because it leads to "% of what." Worse there is no inherent way to know what's the difference between 400% and 800%. To me it is one stop.

The chart is using % as a tag. "Enter by the ENTRANCE sign and leave by the EXIT sign."

They should have used a Greek character because we would not ask the question. We would see them as tags. Problem is, folks might think on character is worse than the other -- and 709 is not worse.

Two meaningless icons would have been better. Open circle and closed circle.

But what to show in the spot meter? Now we need a number! I would have used white letters on black and black letters on white with both sales going up to 100. That would clearly indicate that 255 was the limit for both -- which would be far less confusing.

[Greg Boston]

Originally Posted by Steve Mullen

709 is a CORRECTION -- to match CRT response curve. Use Excel to take linear-light data and convert it to 709. Note where the stops fall. Then multiply the data by the CRT curve and one is back to linear-light.

Yeah, a decision made early on in the days of television. Rather than try to correct the non linear curve in all those tv set crt tubes, they decided it would be easier to apply an opposite gamma curve in the cameras to give a more linear picture result to the viewer.

Sorry, just brought up some thoughts of nostalgia when I read that.

I see what Lee is saying, but the battery life analogy makes good sense.

[Greg Boston]

Originally Posted by Lee Powell

What both JVC and Sony charts have in common is that they're trying to convince us that the curve with bigger numbers has multiple stops of increased DR in the highlight region. But that's obviously bullshit since any tone curve that's exposed at the correct 100% white ETTR level will record all highlight levels below that point as well. In other words:

Dynamic Range starts at 100% white and extends downward to the point where image details are obliterated in the shadows.

If J-log1 truly has "3-stops more DR" than ITU709, it should be able to record 3-stops of additional shadow detail, just above 0% black. I frankly haven't seen anything close to that in my tests of the LS300, I'd estimate it's not even 1-stop better. And that's not because the shadows are too noisy, it's due to both curves encoding the darkest shadow details with only 2-bits of color depth.

What they are really doing in the highlight region is what you touched on regarding the number of bits used to encode the shadow detail. By mapping the sensor's response to light over a longer range of digital values, more highlight detail is preserved, allowing the end user to grade it how they see fit. It gives the impression of increased dynamic range over a curve that is more vertical, where the sensor's output voltage is mapped to a more radical change in digital value.

Technically, they should called it increased highlight resolution, rather than dynamic range. True dynamic range is limited by the sensor's actual ability to respond to increasing light levels before the photosite transistors become saturated and no further increase in light results in increased output signal.

I guess you could call it increased dynamic range if you see it as a crippled dynamic range when using the standard gamma curve.

Another way to explain what I just stated above is this. What is the normal price of my auto insurance? Am I getting a discount for paying in full, or am I being penalized with a higher price for paying it out over time. It all depends on your point of reference as to what 'normal' is.

Going forward, I see higher dynamic range being achieved with the same technique used in stills. Two or images captured in rapid succession(so much so, that normal motion between frames is negligible), with different exposure values. The images are then blended for increased dynamic range. But the hardware will now be able to keep up with this technique at video frame rates and will become a transparent process to the user.

[Steve Mullen]

"Technically, they should called it increased highlight resolution, rather than dynamic range. True dynamic range is limited by the sensor's actual ability to respond to increasing light levels before the photosite transistors become saturated and no further increase in light results in increased output signal."

YES!

That's why I still worry about clipping.

[Steve Mullen]

CAN I USE 8 BIT TO RECORD S-LOG?
ALISTERCHAPMAN

Not, I think, applicable to our LS300 since we get ~11 stops vs ~14 on Sony cameras. Thus, the LS300 needs only compress ~3 stops into 0-255 rather than 5 or 6 stops more.

...

It’s not so much whether you use 444, 422 or maybe even 420, but the number of bits that you use to record your output.

What you have to consider is this. With 8 bit, you have 240 shades of grey from black to super white. ... With Rec-709, standard gamma, on an F3 and most other cameras you get about an 8 stop range, so each stop of exposure has about 30 shades of grey. The stops above middle grey where faces and skin tones are have the most shades, often around 50 or more. Then you hit the knee at 90% and each stop only has a handful of shades (why over exposure looks bad).

When you go to S-Log, you now have around 13 stops of DR (14 with S-log2 and S-Log3), so now each stop above middle grey only has approx 18 shades. Potentially using 8 bit for S-Log, before you even start to grade, your image will be seriously degraded if you have any flat or near flat surfaces like walls or the sky in your scene.

...

I’m sure most of us have at some point seen banding on walls or the sky with standard gammas and 8 bit, just imagine what might happen if you effectively halve the number of grey shades you have.

By way of a contrast, just consider that 10 bit has 956 grey shades from black to super white. ...So when shooting S-Log using 10 bit you have about 73 grey shades per stop, a four fold improvement over 8 bit S-Log so even after shooting S-Log and grading to Rec-709 there are still almost twice as many grey shades than if you had originally shot at 8 bit Rec-709.

...

The whole reason for S-Log is to give us a way to take the 14 stop range of a high end camera sensor and squeeze as much of that signal as possible into a signal that remains useable and will pass through existing editing and post production workflows without the need for extensive processing such as de-bayering or RAW conversion. This isn’t to much of a problem if you have a 10 bit recording, but with an 8 bit recording making it work well is challenging. It can be done, but it is not ideal.