by Art Adams: Sony F5 Dynamic Range

[Chris Hurd]

Sony F5: Dynamic Range, plus interpreting it in Resolve -- by Art Adams

"Sony gives us a lot of gamma table options in the F5 and F55. I recently had a chance to sit down with an F5 and map them all out, and in doing so I learned a lot about both gamma curves and how BlackMagic Resolve Lite handles them.

I had no idea what I was getting into when I shot these tests. Usually I bring clips into Final Cut Pro, export them to Apple Color and capture waveform images from there, but according to Sony there are no current options to import XAVC HD directly into Final Cut Pro 7 or X. I’m not that familiar with BlackMagic DaVinci Resolve Lite, but I learned a lot… while putting a nice dent in both my forehead and the wall over my desk along the way."

Read the entire feature article at Sony F5: Dynamic Range, plus interpreting it in Resolve at DV Info Net

[Alister Chapman]

A good article from Art, I know it takes a lot of time and effort to make and record such measurements. Thanks for the hard work Art...... But, a few small issues with the maths and where Art put's middle grey in the article. The range used for 10 bit video is 0-1023 with black at 64 as Art points out, so in the calculations Art should be using a range of 959 to represent 0-109 IRE and then adding 64 to the result of the calculations to get the actual bit value. But, this is a little academic as the transformation from the sensor output to recorded signal of both standard/hyper gammas and S-Log to data bits is non linear so you cannot use simple arithmetic to calculate bit values for things like middle grey with any accuracy anyway.

For S-Log2 Sony define middle grey at 32 IRE or bit 347 and 90% white as 59 IRE or bit 582. This put's Art's use of middle grey at 40% around a stop too high, so I would add 1 stop from his log over exposure range and subtract one from his under exposure range. Of course where you place middle grey has no effect on dynamic range, just the over/under range although exposing too high with any type of gamma can result in skin tones and mid tones getting a bit compressed which doesn't look nice. Shifting middle grey up and down is how the EI modes on cameras like the Alexa, Red's, F3 and in the future F5/F55 allow you to tweak your latitude up and down.

One area that I also question is the correct placing of middle grey for Rec-709. My understanding is that middle grey is supposed to visually fall half way between white and black. Given that a pure white card will only reflect 90% of the light falling on it and thus "white" would be 90% (anything above being a direct light source), this puts the correct exposure for middle grey at about 46%. To me 40% seems a little low for middle grey and Rec-709 especially as the toe is linear while the upper parts of the curve follow the 0.45 power law. If anything this would tend to put middle grey a little above the middle? Perhaps Art could comment on this?

[Art Adams]

Hi Alister-

Where is the white paper on SLog2? I couldn't find it anywhere, but I do know that they moved something as SLog2 looks more "correct" on set than SLog2 as the shadows are darker, making it look less flat. If they do state that it should be at 32IRE then there should indeed be another stop of highlight latitude.

I have to correct you on middle gray, though. 45% doesn't account for the straight line portion of the Rec 709 curve below 10 IRE. Here's the formula from Charles Poynton:

100 * ((normalized reflectance)^(gamma) * 1.099 - 0.099)

At .45 gamma the result is 41.7, which I round down to 40. And even that seems too bright to me sometimes.

If you can let me know where Sony says middle gray should fall in SLog2 then I'll add a link and a correction. Thanks!

[Art Adams]

I found a link at Abel Cine that seems to confirm 32% for SLog2, compared to 38% for SLog1:

Sony’s S-Log2 and Dynamic Range Percentages | CineTechnica

Going back and looking at my work in Resolve this looks to add 1/2 stop to SLog2's overexposure headroom. I know that black is supposed to be at 64 but in Resolve it seems to sit around 96 and I can't move it without adjusting ped, which defeats the purpose of this particular test, so while pulling black down another half stop would add an additional half stop of dynamic range that doesn't seem to be the way the signal was recorded.

The technique I use when finding the exact clipping point was taught to me by Adam Wilt, who changes the aperture to roll through the clipping point so that he can step through the footage frame by frame and find the exact point where that top step flattens. When I did that here everything changed except where black sits, which is at around code value 96, so I'm going to assume only an additional 1/2 stop of dynamic range from moving middle gray 1/2 stop lower. (Even Sony only claims 1300%, or six stops, not 1950% or 6.5 stops.)

Thanks for catching that, Alister. I've updated the post.

[Alister Chapman]

Thanks Art. I do stand corrected over middle grey and 709. 42% it is. Of course this all assumes the camera manufacturer hasn't massaged their version of 709 to gain a bit better DR. Have to admit I rarely use grey cards with 709 as I'm normally more concerned about my headroom and overexposure than the placement of middle grey. But I do need to keep that 42% figure in mind for my own camera tests. Grey cards are however extremely useful for Log and that's where I use them the most.

The Resolve Data levels or Video levels setting tells resolve whether to scale your output to Video levels or Data levels depending on your chosen output codec. This does have a knock on effect on the scopes as in the case of S-Log it will be stretched to fit the full 1-1023 range for use with computer displays etc or using video the levels is squashed to fit between 64 and 940 or "Broadcast Safe". Sadly there is no "native" mode with no scaling. I have quite a bit of S-Log2 from my F5 that shows a pedestal at about 32 on the Resolve scopes in data mode, so if your set to "video" levels my guess is that resolve is simply adding 64 to the bottom end and then squashing the top down to 940, hence the apparent high pedestal.

[Art Adams]

Yeah, the Resolve scopes were the hardest to get a handle on as I'm not used to reading brightness values as data levels, or having the option of two different ways of mapping the values. I think I chose video levels as data levels were clipping the top end on some of the shots. I'm not sure why that would be, but when I had to reduce gain to see the clipping point I opted for the mode that just showed me the whole range without adjustment.

I called Blackmagic support for assistance and they were fairly helpful, but they also didn't have an explanation as to why they don't support IRE values. That makes it a bit tough to do this kind of testing; historically I'd use Apple Color but as that's end-of-life and can't open XAVC files I'm stuck with Resolve until Sony supports one or both versions of Final Cut Pro.

[Chris Hurd]

Originally Posted by Art Adams

Recently I taught my first lighting class, for Abel Cine, at Sony DMPC in Culver City, California. It got me thinking about how I know what I know about lighting, and why I seem to be able to explain it. One of the worst lessons I learned in film school was three-point lighting.

Read the full article at My First Lighting Class; or How I Learned to Light at DV Info Net