F3 RGB444 and S-LOG tests via GEMINI 4:4:4

[Jim Arthurs]

Hi folks,

Last month I had the chance to shoot some tests with one of the few S-Log enabled F3's in conjunction with a Convergent Design Gemini 4:4:4 prototype. All footage was dual link, uncompressed 10bit, recorded 4:4:4.

Some of this footage was viewed last week at the Cine Gear show in the Convergent booth, and I've received many requests to post more details and share information.

We were hosted by Jon Firestone at his Asgard Entertainment complex in Denver, and our awesome talent for the day was Sara Salazar. Chris Bagley was gracious to provide some behind the scenes stills of the tests that day.

Asgard Entertainment - Colorado's Production Resource
EYEOSAUR.COM


For starters, here's a look at one of the low light level tests we did. Here, the REC 709 gamma exposure was set so that the blue lights behind the screen was full clipping (knee manual, 90 with slope of 10). The rest of the exposure fell where it may below that, and the goal was to see how much grading could be done on the image, and the relationship to the s-log version shot at the same stop.

Glad to answer any questions, and if any are interested, post more tests.

Regards,

[Michael Palmer]

Great to see this Jim, It sure looked good at Cine Gear and I grabbed a copy from Mike. Can't wait to see more. Great job.....

[Alister Chapman]

Thanks Jim.

It's interesting to see how well the Rec709 10 bit grades, other than the ugly clipped lights it looks pretty good. Clearly the S-log has a highlight advantage as to be expected. I'm still keen to see some comparisons between cine/hypergamma and S-Log which should fall somewhere between the Rec709 and S-Log.

[Jim Arthurs]

Thanks guys,

Alistar, I was also surprised how well that the Rec709 could be graded back to normal.

Here is that same test in motion;

http://ftp.datausa.com/imageshoppe/o...9_VS_S-LOG.mp4

And you folks dying to get your hands on actual RGB 444 material, here is a zip file with the non-graded .dpx files (Rec709 and S-Log), for anyone to take a stab at;

http://ftp.datausa.com/imageshoppe/o..._VS_REC709.zip

I'm not a colorist by trade, so others might enjoy getting a feel for how this footage manipulates, and get better results.

Bringing the RED into the equation;

In addition to the F3/Gemini444 grabs, that zip includes a clipped down single frame .R3D file from the RED camera with latest MX sensor that was shooting the same scene. We switched out and used the same Sony glass at the same stop on it as well to get a feel for how these two systems perform under low light conditions.

Using the clipping blue light as a guide, we all felt the RED would have enjoyed having the lens a half stop or more open, so the compensation was a higher ISO rating (1280 vs the F3 at 0db 800). Of course, changing the RED ISO is simply meta-data that can be changed in post, or ignored, but gives the grader a starting point.

I provide the .R3D instead of a scaled down .dpx with a grade applied, because RED has always suggested that any tests like this be done with the full resolution native files. Now that you can actually create single frame raw files, this is an easy process. Of course this means you'll have to download the latest REDCINE app from their website, but it's free...

One final note, these tests are under 3200K lighting.

Next up, what happens with this same low light setup if you decide to stop down from 2.8 to 8? What will S-Log combined with 18db of gain look like? Is it still a usable image?

Regards,

Jim Arthurs

[Jim Arthurs]

Okay, on the topic of white balancing when using S-Log... I haven't seen a single post anywhere about this from the early users on any forum, but at least on the F3 I'm working with, using an S-Log upgrade dating back to NAB, all white balance functions on the camera are disabled. Trying to toggle the white balance pre-set switches or pressing the front button simply gives an error message.

RGB 4:4:4 will white balance, but not when the camera is set to S-Log.

This very well could be a beta firmware bug, but if it isn't, everyone should start to prepare for what this means in terms of usage. When S-Log is enabled, white balance defaults to 3200K, and the only camera internal change to this is when pressing the "5600 CC" switch on the side of the camera, which balances for daylight. This apparently is fully electronic, as no change to the stop is noticed. Any other manipulation of white balance will have to be done via external lens filtration.

I bring this up prior to releasing some green screen still grabs in a day or so, where you will notice slight off color to the grabs (Approx 4700K fluro lighting, could only choose to balance at 5600K on camera). Of course these grade right back quickly and easily to match "proper" white balanced non S-Log material.

Maybe Andy from Able or someone with access to a recent upgrade could comment?

Regards,

Jim Arthurs

[Nate Weaver]

I played with an S-Log enabled camera (actually a couple) at Cinegear last week, and toggling between 5600 and 3200 presets seemed to be working. I don't recall seeing any error messages. It was difficult to see if it was changing because I think the WB ramp speed was set to slow, and S-Log is pretty desaturated as it is.

In my experience with Red footage, where all white balance correction is effectively done after the image is digitized and compressed, that method works just fine too, if the F3 S-log did happen to lock in a non adjustable white point. Just as long as things are 10-bit or better, correcting in post is not such a big deal.

[Nate Weaver]

Jim,

Something is up with the noise in the S-Log DPX:

I haven't seen noise like that out of my F3, even with +18 gain I*don't think.

Have any idea what happened? It doesn't make sense...

Snapshot below is the DPX alone, without any manipulation or LUT applied.

[Jim Arthurs]

Hi Nate. I'm finding that S-Log material is noisier in general. I double checked to make sure I zipped up the 0db frame, instead of the 18db one (which is coming... ;).

Bear in mind that shot is under tungsten, and like the RED MX, I think the F3 sensor favors daylight balance, so there's more noise in the blue channel.

Second, the effective ISO is, what, 1600 compared to normally 800?

Do compare it to the .R3D extract frame in the zip, I think you'll find the S-Log to be better in all color channels, and that's factoring in the 2X down-sample advantage the RED gives..

The next video I will show will be exactly that, split screen of full resolution showing F3 S-log at both 2.8(0db) and 8.0(18db) vs the RED MX sensor at both of these stops...

[Nate Weaver]

I would maybe agree, except I have some of my own S-log from Cinegear (daylight) that is just as noiseless as anything else I've seen out of the camera at 0 db. I'm very familiar with the noisy blue channel issue on Red/other cams at 3200... But the gap seems too much here.

So not sure what to think...

[Jim Arthurs]

It's always possible the lack of white balance functionality (if unique to this camera) is tied into some over arching issue with this firmware beta, but I don't think so.

My understanding of the way S-log works on this camera is that it's taking the full 12bit linear range of the sensor, including all the noise floor, and mapping it into 10bits. Everything S-Log by nature should need some grade just to pull the blacks down to tuck that noise away. Maybe I'm wrong...

This was an extreme test by nature, by setting the exposure of the blue backlights near clipping, and letting all the other areas fall where they may, under tungsten.... again, if you load up and look at the .R3D file and compare channels, you'll see better performance from the S-Log than the MX sensor under the exact same conditions. And just wait until we stop that lens down a couple more stops and re-test.. :)

Here's a zip with DPX (and .R3D) of near daylight green screen, with no attempt to stress the images. See if this compares to your expectations or tests...

http://ftp.datausa.com/imageshoppe/o..._VS_REC709.zip

Also, I would love to see your footage so we can start generating some data points on this as part of the overall education benefiting everyone, so if you can export out a frame or two...

Regards,

Jim Arthurs

[Nate Weaver]

Here's a 1:1 pixel grab from a cam that was at Cinegear.

This is recorded 8 bit to SxS, 0db, no LUT or anything applied. It's practically noise-free.

Again, I know you are right in that 5600K is going to have less noise than 3200, but...

[Nate Weaver]

Originally Posted by Jim Arthurs

See if this compares to your expectations or tests...

http://ftp.datausa.com/imageshoppe/o..._VS_REC709.zip

Yeah, that S-Log frame is looking a lot better.

Can I ask how these DPXs are being created? The black levels between the first S-Log frame and the second are completely different.

The black levels in the second frame go almost down to 0, while the levels in the first frame are elevated, as I'd expect.

BTW, Thanks for entertaining all this spanish inquisition on your frames. I hope you don't feel I'm shredding your hard work, just trying to figure out why that first S-Log frame looked nothing like I'd expect.

Originally Posted by Jim Arthurs

My understanding of the way S-log works on this camera is that it's taking the full 12bit linear range of the sensor, including all the noise floor, and mapping it into 10bits. Everything S-Log by nature should need some grade just to pull the blacks down to tuck that noise away. Maybe I'm wrong...

Yes, but my experience is that neither Red, nor Alexa, nor F3 in reg mode really need this as their noise floors are so great looking already (at their rated gains). I find that technique to be the refuge of the perpetually noisy, and I'm not naming names.

[Jim Arthurs]

Nate, all the .dpx files were pulled from Uncompressed RGB 444 QT files in a 16bit AE project, I do believe that the Cineon export settings may have been on full range once. Truth told, many methods were used to pull the .dpx files with wildly different results. Only the Digital Fusion export workflow seems to be predictable for me 100% of the time, probably due to my comfort level with the app.

The Gemini will directly record to .dpx image sequences at initial release, which will eliminate a lot of the transcoding issues.

As to any noise... I will say that the F3 we have shows noise floor noise on high end scopes when in S-Log mode. Your sample looks cleaner than I've seen so far, especially in the blue channel. I'm thinking a re-apply of the final firmware may be in order... :)

Regards,

Jim Arthurs