Gemini 4:4:4 frame Grabs

[Tom Roper]

If I am to understand, S-log is the equivalent of camera raw output for video, and the Gemini is for capturing the S-log feed in a manner that preserves maximum chroma detail for subsequent grading in post. Accordingly the S-log grab appears very flat and compressed, because you intend to stretch it back out in post.

With the cinegamma or standard gamma the image is burned in, so if it needs further correction in post, there's less to work with. So if like Doug you never make a mistake, always get it right, you don't need S-log.

I usually get it right too, but then why with dslr stills do I prefer shooting raw? Must be because it's pretty good insurance for capturing the scene in the entirety.

As for the opinions about certain looks, Piotr prefers cinegamma and Doug is highly critical of clipped highlights while accepting of crushed blacks. The latter makes little sense as it seems likely to displease as many who prefer shadow detail and don't care as much about clipped highlights. Since it's just opinion, mine is for std gamma R709 with well chosen knee settings to best preserve highlight AND shadow detail, followed by post grading only if necessary.

That said, if I had S-log and the Gemini 444, that is what I would use because you could ultimately achieve any look at any time through post edits, where if you made a mistake with your picture profile in-cam, that look will be burned-in.

[Peter Moretti]

S-Log isn't raw in the way that a DSLR, Red or F65 can shoot RAW. S-Log has R, G and B data for each pixel, RAW has only an R, G or B value for each pixel and has to be reconstructed in post to create an image that's not a mosaic of colors.

What S-Log does is allow the full dynamic range of the sensor to be recorded. This is accomplished by using a logarithmic curve for the encoding that decreases sensitivity as the image gets brighter. This is similar to how film records light and how our eyes see light.

What frequently gets lost in the S-Log discussion is that many modern sensors clip well past where the recorded image clips. This is for two reasons: 1) Sensor dynamic range is wider than what you could properly record w/ many codecs, e.g. XDCAM-EX. 2) Most video applications don't need or even want the extra dynamic range b/c it surpasses what most broadcast systems are designed to handle and display.

So in truth, if the highlights roll-off gracefully, the extra dynamic range is not needed or even desirable for a lot of the jobs that video cameras are used. But for cinema type applications, the extra DR can be very helpful, and S-Log fits well into existing film workflows. A main need for extra DR is time of day color correction. This is esp. true when shooting outside b/c the light is constantly changing, but shots from the morning may need to match those from noon and later. And of course, there is day for night shooting. And there are times when indoors that you actually want to see what's walking past that window facing outside.

But Doug is right in that the end product is frequently going to have less DR than S-Log provides, so it can be a lot of work for not much benefit, depending on how you work what you want your images to look like.

Hope that all accurate and helpful.

[Tom Roper]

Thank you for that excellent explanation.

[Billy Steinberg]

Since the crux of this discussion is about dynamic range, I thought a graph of a standard (HiDef) gamma curve versus the Hyper (Cine) and S-Log gamma curves might be interesting so I attached one at the bottom of this message.

Note that my niche is live broadcast video. The cameras I use have separate red, green, and blue imagers, not one sensor with a Bayer filter. I shade and paint (color grade) the pictures in real time; it's usually not done in post, unless I screw up. 90% of the time I use Sony broadcast cameras, and the curves in the graph below are the gamma curves that Sony uses in all their cameras (both single chip and three chip). Different brands of broadcast cameras have different "standard" and Hyper (Cine) curves, but the principle is the same. The "standard" HiDef gamma curve follows the Rec709 spec, and all camera brands will have at least the standard Rec709 gamma curve (or the 240M curve, which is very similar, differing mostly in the color matrix, not the gamma curve). Sony cameras have six "standard" gamma curves and four Hyper (Cine) gamma curves available for selection. In the "standard" gamma curves, Sony also lets you gently alter the standard curves with a master step gamma control (.50, .45, .40, .35, with the "standard" being .45, which leaves the curve tracking the Rec709 spec). In both the "standard" and Hyper (Cine) modes, Sony also gives you what they call "Black Gamma" which allows you vary the curve in a ±5% range at either 5%, 15%, 25%, or 35% input level. All of the gamma adjustments except the master step are individually available in red, green, and blue (as well as a master control that varies them all in sync). There are many knee controls available as well, but although they are relevant to dynamic range compression, this message is going to be long enough as is...

Note the difference in the graph below between a standard HiDef gamma curve (labeled R709 in the graph), and any of the Hyper (Cine) gammas and the even more severe S-Log curve. One can squeeze two to three stops more at the top without losing info the sensor is capable of resolving. Note also that the curves differ in the low and midrange as well. When one is "messing" with the gamma curve of a camera it's not just the "whites" (70%-100% input level) that are changed!

The whole purpose of all of this is to squeeze the dynamic range of life into something that we can view on television. Our eyes/brain already limit life's dynamic range, but film and video have a much more limited dynamic range. You can still squeeze more dynamic range out of film than video, but not a whole lot more, and film's advantage these days may lie more in the graceful way it handles overload (too much light) compared to video. (And maybe that film's grain is black and video's noise is white).

The camera sensor(s) we use these days can already capture more dynamic range than we can distribute on video. But simply altering the gamma curve of the camera to fit its total dynamic range into what can be recorded makes for a really ugly picture. The S-Log curve does just that. (And if you've ever looked at uncorrected S-Log video, you know how ugly it looks). Unfortunately, if you just apply a reverse S-Log curve to the video (via a LUT in production or color timing in post), it still doesn't look great. Much better than raw S-Log, and good enough to judge exposure, but not nearly as good as if you trim the gamma curve to fit to the picture you're grading.

To really take advantage of the sensor's expanded dynamic range, you have to understand how gamma curve changes effect the picture you're shooting. You can do it when you shoot or when you grade the picture in post. If you don't record a S-Log version when you shoot, your ability to further grade in post is severely limited compared to what you can do if you have a S-Log version. On the other hand, if you're shooting a show that's being broadcast live, or a show that won't be post graded, you can reap great benefit by setting the gamma curve in the camera appropriately for the content you're shooting.

When you've been playing with gamma curves for a while, you'll start to "just know" what camera settings benefit different shooting circumstances. While both shooting a classical orchestra indoors and shooting Eric Clapton in an open outdoor arena both benefit from using a Hyper (Cine) gamma table and trimming down the black gamma, the gamma curve used and the black gamma settings are completely different. In the former, one is generally looking to remove the bleaching effect on fingers, noses, cheeks, and bald heads of what is almost always too much top light (and preventing the sheet music from becoming a nuclear rectangle of white light); in the latter more severe correction is needed to deal with an audience in noon day sunlight and a stage with performers that's many stops less bright, both in the same picture. You set the iris to put the desired part of life's dynamic range on the less dynamic sensor, and you set the gamma curve to put the desired part of the sensor's dynamic range on the less dynamic video. With the iris you can only move the top and bottom of the range, together, without altering the dynamics of what's in that range; with the gamma curve, you can also alter the dynamics within the range you are recording/broadcasting/viewing.

The graph below gets much larger, with much more detail, if you click on it (or download it). The HG1, HG2, HG3, and HG4 curves correspond to the four Sony Hyper Gamma curves. In the broadcast world, we usually set our white clippers to 102%, and we assume that somewhere along the line they'll probably be clipped to 100%. HG3 and HG4 were calculated by Sony to work properly if one opens up the white clippers in the camera to 110%, and expects that the signal won't be clipped lower until after the grading is done in post. None the less, HG3 and HG4 happen to work fine if you set your white clips to 100%-102% anyway, and are less severe than HG1 and HG2. HG3, for example, works really well for classical orchestras.

By the way, if I'm not mistaken, the "S" in S-Log stands for Sony...

Billy

[Doug Jensen]

Billy, fantastic post! I've never seen such a thorough and easy to understand explanation of the gamma options -- especially since it comes from someone with actual hands-on experience on real shoots instead of theory or charts. Sounds like you're exactly the kind of VC I used to love learning from back in the days when I was working on trucks. Thanks for your input.

[Dan Keaton]

Dear Billy,

Great Post!

Thank you very much for taking the time and effort to post your experience and knowledge!

[Alister Chapman]

Where S-Log really comes into it's own is in situations where the DoP needs to capture as much dynamic range as possible so that the down stream colourist can manipulate the image in post production. Compared to REC-709 (the standard to which most HD TV's are adjusted) the shape of the curve means that there is a lot of gain/lift in the shadow and low key parts of the image with less and less gain as you go up the curve, reducing to almost zero close to where the sensor/DSP itself reaches it's limit.

The benefit that this brings in post is that generally speaking, low key areas will be reduced in brightness to increase how contrasty the image looks. The reduction in low key level will also decrease the appearance of noise in the image. Well exposed S-Log will require minimal gamma adjustment to skin tones and natural textures such as plants and foliage. Then highlights can be tweaked to suit the mood of the shot. As our own visual system is naturally less acute to subtle highlight and extreme light levels, large inaccuracies and inconsistencies in these areas will go largely un-noticed.

Cinegammas/Hypergammas share some of these characteristics, however they have very similar gain levels to Rec-709 at the lower end of the curve so low key parts of the image look more natural without grading. If you did have the same lifted shadow areas, recorded with just 8 bit data, you would very often run into banding and quantisation noise issues when pulling the levels back down. With a live camera feed, which is normally 10 bit or uncompressed component, you don't have the added problem of encoding and compression artefacts to deal with.

Simple, quick turnaround projects, run of the mill TV work etc will still in most cases be best dealt with by carefully setting up the camera with a conventional gamma and only minimal post production grading. However S-Log brings clear benefits for workflows where the image will be manipulated in post, not just in terms of the increased dynamic range but also the way it grades cleanly.

[Peter Moretti]

Right (of course) as what can be noticed in the graph is that HG 1 and 2 flatline at the right side. In contrast (pun intended) the S-Log curve is still sloping upward, recording meaningful values.

[Billy Steinberg]

Originally Posted by Alister Chapman

Cinegammas/Hypergammas share some of these characteristics, however they have very similar gain levels to Rec-709 at the lower end of the curve so low key parts of the image look more natural without grading.

Actually, if you look more closely at the curves, you'll see that S-Log is much closer to Rec709 in the dark areas (below 40%) than any of the Hyper gamma curves. That's one of the reasons why pulling down the black gamma when running Hyper gamma makes for a much nicer picture.

If you did have the same lifted shadow areas, recorded with just 8 bit data, you would very often run into banding and quantisation noise issues when pulling the levels back down.

While one might run into banding/quantization when grossly stretching out a section of video, the same doesn't happen when shrinking it (pulling the black levels down). In fact, the opposite happens. And the cameras I work with have 12 (and occasionally 14) bit data, not 8 bit. We are talking about Hyper (Cine) gamma here, which happens in the camera.

Simple, quick turnaround projects, run of the mill TV work etc will still in most cases be best dealt with by carefully setting up the camera with a conventional gamma and only minimal post production grading.

As I said in my post, I think that S-Log has definite advantages when one is shooting with post grading in mind. Hyper (Cine) gamma can be used to GREAT advantage in simple, quick turnaround, run of the mill TV work, and for what it's worth, the last two Emmy Award statues I won were for shows where I used Hyper gamma; neither of the shows were simple, quick, or run of the mill.

Billy

[Peter Moretti]

Billy,

Thanks MUCH for your excellent contributions to this discussion. I just want to assure you that after having read Alister's posts and seeing his tests over the years, there is nothing egotistical about the gentleman.

/resume normal programing :)

[Piotr Wozniacki]

Originally Posted by Peter Moretti

Thanks MUCH for your excellent contributions to this discussion. I just want to assure you that after having read Alister's posts and seeing his tests over the years, there is nothing egotistical about the gentleman.

I've only been following this interesting thread passively (as neither my technical knowledge, not the equipment I can afford using do not entitle me to take an active part in it) - but I must say that with the above statement, Peter is absolutely right!

[Doug Jensen]

Originally Posted by Billy Steinberg

Hyper (Cine) gamma can be used to GREAT advantage in simple, quick turnaround, run of the mill TV work, and for what it's worth, the last two Emmy Award statues I won were for shows where I used Hyper gamma; neither of the shows were simple, quick, or run of the mill.
Billy

Absolutley True. This is exactly what I have been trying to say for many months but a lot of people seem totally fixated on over complicating the issue of getting great results from the F3. I would even go so far as to say that I can achieve nearly the same results for broadcast (8 bit delivery) with Cine (hyper) gamma -- as can be achieved with S-LOG. Okay, for heavy grading, CG, theatrical delivery with a 10-bit projector or transfer to film, S-LOG probably makes a better choice. But that's not my market, and I suspect not the market of most us hanging out here at DVinfo.

And the camera's normal paint menus can do what they do with no workflow hassles, no expensive firmware updates, and no expensive external recorders. The F3, right out of the box, has some amazing capabilites built right into it if someone cares to take the time to paint it properly and then (very important) expose correctly. Yes, it is less forgiving than S-LOG, but I prefer not to shoot sloppy. I approach all my shooting as if I was shooting for a live broadcast on everything I capture. The picture looks great on the client monitor while shooting (the built-in LUTS do not!!) and the footage is immediately ready for editing.

For those who say differently, I'd love to see some examples of your work. No frame grabs, no charts and no table-top still lifes. Where is the real-world video that proves what you're talking about? I've seen Billy's work many times and I know he can walk the walk.

BTW, I think this thread should be moved to the F3 forum since it has very little to do with Convergent Design.

[Chris Hurd]

Moved to Sony F3 forum from Conv. Design, per suggestion. Topic deals more with the F3 than with Gemini.

[Brian Drysdale]

Originally Posted by Doug Jensen

And the camera's normal paint menus can do what they do with no workflow hassles, no expensive firmware updates, and no expensive external recorders. The F3, right out of the box, has some amazing capabilites built right into it if someone cares to take the time to paint it properly and then (very important) expose correctly. Yes, it is less forgiving than S-LOG, but I prefer not to shoot sloppy. I approach all my shooting as if I was shooting for a live broadcast on everything I capture. The picture looks great on the client monitor while shooting (the built-in LUTS do not!!) and the footage is immediately ready for editing.

I don't know about the F3, but you could get a great range of looks in camera with the Panasonic SDX 900's paint box. You could see instantly how the set colours were working on the monitor rather than waiting for post.

[Doug Jensen]

Brian, it's the exact same way with the F3 and all the other XDCAM cameras. Dial in your "look" in-camera and then lock it in and start shooting. No grading or correcting necessary -- unless you're looking for an extreme look or have to push the footage through a lot of CG. Of course, if you can't decide what look you want, or can't be bothered to nail the exposure properly, then that is where S-LOG comes in handy. There aren't that many shooting situations I run into where a couple extra stops of dynamic range is really going to make a big difference vs. all the other 3rd party gear and workflow hassle that are involved to work with S-LOG. That's my opinion today, but I reserve the right to change it after doing more real-world testing.