A Brief History of Quality: Comparing BRAW 3:1, Q0, Q5

Here is a short review of real world quality when recording BRAW 3:1, Q0, and Q5 on the BMPCC4K. It’s designed for pixel peeping which you can do by going to the boundary of each clip and toggling between two quality options. I created a video in both 3840x2160 and 1920x1080 although my current projects have 2K/HD deliverables regardless of the capture quality options. You’ll see the details for the clips and can make your own conclusions in terms of how you intend to shoot your clips.

I went into this with the assumption the Q0 would produce the best results. I have upgraded my media capture cards to be able to handle Q0. So I’m armed and locked for Q0.

This subject might seem an odd test, but when you study it carefully, you’ll find it’s very revealing. There are many items in motion and some static items, and the wind and sun inevitably vary in an outdoor shoot. But there’s a wealth of individual leaves and rocks upon which to make judgments about the perceived quality even when massaged to h.264 and then uploaded to the cloud for your viewing pleasure.

Here are my conclusions:

In Resolve 16b6, using Resolve Colour Management, for these complex clips, you can see BRAW 3:1 only has the advantage of predicable storage requirements as I found here Constant Quality Q0 produces the more satisfying image generally.

Q0 versus Q5? Of course Q0 must be better because it’s a lot higher data rate. The design goal of the ‘Q' is a quality image and how could Q5 match Q0 with so much less data? In a word: it’s Blackmagic. Sometimes some elements in either Q5 or Q0 are ‘better’ than the other; it’s not a race in which one runner is in the lead throughout the race. Q5 makes some sacrifices that you can see, but it seems to have some less understood running style and I have to admit in this video, it emerges victorious, not only in fidelity but also edge sharpness. Astounding what has been created here. It contradicts my prior conclusions from some of the static demonstrations I’ve seen before. Throw away test charts that don’t move or respond to dynamic changes in light. I hate to admit I may have been wrong.

Please don’t rush out and conclude Constant Bitrate 12:1 is great; other clips I shot immediately showed it lacking, but Q0 is not like 12:1 in a complex scene. It can be more like 7:1 which isn’t an option, 8:1 may be fine—I’ve tested them all of course. But today is all about the race for quality. And Q5 is able to win this race now that Usain Bolt (represented by uncompressed raw fury) has retired.

I understand this might be different if you were generating deliverables for cinema, but this test assumes television or web consumption. When i’m producing for a cinema distribution, I’ll render appropriate deliverables and review.

A shoutout to Jamie LeJeune who first suggested that the benefits of shooting Q0 may not be worth the increased costs for media and storage. In trying to prove Jamie wrong, I think I’ve convinced myself for most everything I shoot, Q5 will be the best value proposition. The only exception I think may be the narrative movie, but at this state of my finances, trying to afford the newer Mac Pro, I may use Q5 for that film too. I have another huge project potentially and I know Q5 will be the best choice for it as it will require an awful lot of raw storage and I now know that Q5 isn’t a ‘cheap’ solution, it’s quite capable of great images with a 12bit log base.

Great test! Thank you for sharing Rick. We geeks always love a good comparison video

I also used mostly q5 in the beginning, but found some static scenes with less motion to be even worse than 12:1 for some odd reason. It became a bit mushy, worse than prores lt. Could be that q5 goes to low when it feels to... introducing compression artifacts. 8:1 I find as a sweet spot in many ways but it depends on what I’m shooting. So for a tripod shot with a talking head and static background I find 8:1 or even 12:1 better than q5.

Yes.. unfortunately it seems like the q5 have a tendency to use lower bitrate than it should in some static scenes, IMO. There should be something between q5 and q0 in that regard.. maybe they will implement it or update the q5 to be more robust in the lower end of the scale. In the upper end it seem really good. 5:1 also seem like a good choice.

I have settled on Q5.

Thanks for the feedback, folks. I didn’t think to test a mostly static and simpler scene, so looks like I need to watch for that. I’m not opposed to using 5:1 if it gets around the issue of Q5 dipping too low at times. Complicated stuff seems to defy the usefulness of rules of thumb. The other option for me could be to shoot Q0 at times where Q5 may not be the best choice for the reasons you’ve given. On a very complex motion scene Q5 is amazing. On a simpler situation Q0 might be the better choice and the Q0 bitrate will be at the low end which is somewhere akin to 5:1 (need to check). Since I’m shooting with terabyte sticks, the media storage isn’t so much a concern.


Sent from my iPhone using Tapatalk

Seems to me that, as someone mentioned earlier, 8:1 is the best compromise since it never dips too low and it's only slightly smaller than Q5 at the top end.

Turning around a lot of work as we do here, the ease of working with Q5 in the field and in edit is a no-brainer.

It just looks great and edits and CCs fast.
The fact that you can get 40+ minutes on a Sandisk Extreme SD card is pretty silly for $35 a card. It's like disposable media.

Amazing how Braw/Q5 changed our workflow since it was launched.

I also used mostly q5 in the beginning, but found some static scenes with less motion to be even worse than 12:1 for some odd reason. It became a bit mushy, worse than prores lt.

Do you have a file showing Q5 looking "mushy" that you can share?

I've shot well over a hundred hours of footage in Q5 BRAW at this point in a wide variety of scenes on multiple BMD cameras and I have never seen it look mushy due to compression.

This video looks at a statis scene in Q0 and Q5. This scene is simpler than a moving scene and requires less data. The concern is that a simple scene may push the Q5 bitrate too low and artifacts may result.

I found slight differences where Q0 did give a better result, but not sure i found real artifacts. Sometimes an edge as different, sometimes smooth colour gradients were different, but nothing stood out that might be a distraction in an actual motion video with a story happening.

The last two clips are the most interesting to me as I’m trying to get more day-for-night experience. i underexposed the original clips and then in post took two paths to make it look like it was part of a night scene without actually changing the colour. It did not matter if I applied either technique to either Q0 or Q5 as the results were visually the same. You’ll see in the first clip, I reduced the Exposure in Camera Raw and increased the Contrast in the Primary adjustments. in the other clip I reduced the Exposure and did not alter contrast. I’d appreciate hearing from folks if they liked the result of changing one;y Exposure to preferred changing Exposure and Contrast.

Thank you!

[CaptainHook]

I also used mostly q5 in the beginning, but found some static scenes with less motion to be even worse than 12:1 for some odd reason.

Q5 can go up to 20:1 or even higher as indicated on this chart showing "typical" data rates:



(550 / 27)

https://www.blackmagicdesign.com/produc ... ckmagicraw

Chart is from the above link.

Wouldn't it make sense to introduce a Q3 mode as another option between Q0 and Q5?

Thanks for including the chart in this thread, CaptainHook. As you can tell here, we all have our estimates of what might be best practices for a given project or even just a scene. If you have any insights, you’d like to share, we’d love to read them.

Trying to make assessments by pixel peeking doesn’t always translate into apt conclusions for motion video. From what I’ve seen, Q5 sacrifices very little for potentially huge savings in dailies and longterm storage. But Q0 has its strengths too. It’s each person’s choice.

The interesting point here is that there are differences between motion and static. AFAIK, technically BRAW is an I-frame only codec. It should be, given the low load when decoding.

So this must be perceptual, right? It is absolutely possible if we think of opto-chemical film, where a single frame showed pretty bad grain when seen as a still. When the film was in motion the human brain did quite a bit of integration, since the grain structure was completely random in size and position over time.

If something similar is happening with BRAW, BM must have done some serious black magic while encoding!

Wouldn't it make sense to introduce a Q3 mode as another option between Q0 and Q5?

It would make sense if that kind of in-between level matters in actual use. But does it really?

Unless I was shooting VFX plates or super wide static shots where I planned to punch in by a large percentage in post, cases where I might need to use Q0, everything else is in motion or it's shots of people/faces where we are either dealing with motion blur or we do not want to see every individual pore anyway. And, in all my shooting in BRAW, Q5 has been more than sufficient for the vast majority of shots.

When BRAW first dropped, I was worried that Q5 was going to be too compressed until I started testing it, and then began using it more and more for real projects. Over time, I had to just accept the reality of what I saw in the footage -- it's the prefect codec for 95% of what actually gets shot.

So, my question (asked respectfully) is: Do we need a "Q3" in actual real world use cases, or is it really just that it's like a security blanket on the assumption that Q5 is "too compressed" for most use cases, without any actual evidence to back up the request?

If anyone has actual footage (from real scenes, not contrived tests) to show where BRAW Q5 compression has compromised the image in some way, it would be very useful to see.

In the meantime, I'll be rolling in Q5 and saving money on hard drives

In my tests, I was looking for some of the ‘traditional’ problems such as moiré and fixed pattern noise. My last test posted today was shot at 3200 and 1250 intentionally underexposed. You can see difficulties with mouse on individual frames, but seeing it as a video, it looks better. I can’t see the familiar FPN I thought would show up on the underexposed chair back. The first video in this thread includes a roof that I thought could show moiré near the top but I don’t see any.

You can tell the URSA Mini 4.6K has a real competitor in the BMPCC4K in several use cases. If the BMPCC4K had 15 stops... and a 2K recording option... and support for widescreen... and an EVF! It could easily be you main squeeze.

In reality, if I want to deliver 2K, I can shoot 4K in Q5 more easily than shooting CinemaDNG 3:1 in 2K on the Mini 4.6K and with a wider angle of view.

In reality, the support for widescreen is almost there as I can mount my 1.33x Anamorphot and still make sense of the squeezed UHD image on the monitor and that provides me with economical widescreen.

In reality, 13 stops is quite decent and encourages one to be more careful at times with managing light levels and good practices.

In reality, the lack of an EVF can be overcome with the help of third-party vendors. It’s not like the BMVF was joined at the hip to the Mini 4.6K/Pro/G2. It’s just an option and there are other options available today.

It’s so easy to be your main squeeze now. Having BRAW makes a world of difference. I’ve been using ProRes a lot after initially shooting CinemaDNG raw. The BMPCC4K is bringing me back to shooting raw. The constant bitrate option is there as always, but the constant quality options are irresistible to some. I used the BMPCC4K for a client shoot but it was ProRes. I haven’t shot BRAW Q0 or Q5 for a client yet, but I’m confident that will occur in August.


Sent from my iPhone using Tapatalk

Rick, thank you very much for the time spent and sharing this test!

Jamie, I’m traveling at the moment and don’t have a chance to upload anything. See if I can upload when I get back.

My finding was some loss in detail where the texture is complex(static) introduces with something that looks like mosquito noise. I didn't see this in any of the other compressions.. Same scene at 12:1 was better. I have to add that this is by pixel peeping. It’s not a “omg!” moment. And most people would probably not care. Since 12:1 is already quite low when it comes to data rate, and predictable in size, I find it best for talking heads combined with tripod. If you use q5 handheld/paning etc it seem great.

Captainhook, thank you for the feedback. Is there any plans for update regard the braw that you can tell us about ?

... My finding was some loss in detail where the texture is complex(static) introduces with something that looks like mosquito noise... I have to add that this is by pixel peeping... I find it [12:1] best for talking heads combined with tripod. If you use q5 handheld/paning etc it seem great.

That’s helpful in terms of where I need to be careful about Q5 and could be an issue then in any narrative film. In the last test I uploaded I did notice some busy noise, but I put that down to my use of ISO 3200 with exposure set low as I’m trying to prepare for a nighttime (illusion of lights out) bedroom scene! And there’s no noise reduction in my tests.


Sent from my iPhone using Tapatalk

The interesting point here is that there are differences between motion and static. AFAIK, technically BRAW is an I-frame only codec. It should be, given the low load when decoding

When things are in motion, they blur, and blurry detail and out of focus detail compress more easily. Static shots will have more sharp detail and thus require a higher bitrate.

So it's not strictly motion that changes things but the amount of detail.

Definitely. When you wait for the estimates regarding space left, as done by the camera after a few seconds, you can see that lots of small, contrasty details make all the difference when using constant quality.

Good point and agrees with Tim S comment in the other thread that lots of high contrast detail will require the highest data rates from Q0. Fortunately interior narrative films are less likely to push Q0 beyond 3:1 data rates. I did have portions of the shots of leaves that were high contrast edges where the camera was stationary, but now I understand when to be cautious.


Sent from my iPhone using Tapatalk

Hi Rick, thanks for this. I was going point out the complexity issue to. In highly compressed cameras a trick was to increase the depth of field to improve compression. Focus blur didn't help compression. A more complex scene with tax out the lower data rate more (same compression routine) and motion blur also interferes with the predictable spread of light and color over an object. So, a complex detailed scene with motion and focus blur might trip things up more. But, between taking shoots on a windy day, the scene complexity can change dramatically from one shoot to the next.

For my own camera testing, I was going to use the following (you can use lit pictures or complex scenes ): a HDR display with plenty of light and rec2020 color range and accuracy. You then can display suitable complex frames and compare the data rate. However, you are going have to match exposure top to bottom.

BTW, it also means you have a 100% known complex data to compare the compressed frame to gain the % difference, which going change frame to frame with noise, but you get an idea of lowest to highest % similarity over multiple frames. Say 10-1000 frames will give you a very good idea. However, you have yp use a filter to block non primary light glow to get the best match to the file data. You can Aldo lower brightness of the scene (through computer or display, or camera, but mathematically processing the scene in high bit depth, and in camera lowering exposure (to test that) might he best.

Especially now with the P6K, it's nice if Q5 proves to hold up for most use cases. Files do get bigger on that camera.

That said, I feel a Q3 that goes from 5:1 to 12:1 (known entities) would be a sweet spot for many. When you see the compression to towards 20:1 a knot starts to build in the stomach. Especially since there is no control over how the recording happens.

On the other hand, I really hope some critical reviewing has gone into this on BMD's side and that those compression ratios were chosen for a reason. Information is our mutual friend. I'd love a white paper.

I just got my P6K yesterday. I'll do some tests too and post the results if I find something notable.

I'll tel you what, try the above with a simple checker board pattern of light grey on white background dots to match dot per camera sensor pixels precisely at uhd, and tell me what you get? Then try that with cdng to show how god braw is?

I forgot, as Braw is really 4:2:0, a 3:1 times smaller file than bayer would really be 4.5:1, 6:1 9:1 etc. You really want 2:1 Braw then.

I became curious and shot this simple (static) scene as a starter (compressed jpg reference here):



I was thinking that maybe some of the lower contrast texture of the wood would get lost with more compression. Not really. Oh, and there is some minimal lighting and overall contrast variance due to the available light setup with large windows and moving clouds outside.

I shot on a P6K with a Sigma Art 50mm at f4.8. So, sharp where it needs to be, with some fall off to see how that looks.

Uncompressed .pngs can be found here:

https://drive.google.com/drive/folders/ ... sp=sharing

The frames are grabbed stills from Resolve exported to png, so they are all the same size even if the underlying braw was recorded at different rates. One single 6K png is around 63MB, so there's that.

I guess the true test is to record many scenes (motion) over time and do a long term evaluation. I'm thinking that with the P6K, I will start with Q5 and do a lot of tests to see if it ever goes below my personal threshold—I'm doubting it will.

Yes, zero compression is like a "baby comforter" but I like a more practical approach most of the time.

Andree, I think many would like to shoot an in-between quantization. I’d like to see a value that results in bitrates very close to fixed rate 3:1 to 8:1 since these are the fixed rates I would chose if the Qx were not available. That might be a Q2, not Q3, but I don’t know how to do the math you can use to determine the quantization value.

Q0 gives ranges from 2:1 to 5:1.
Q2 might be from 3:1 to 8:1.
Q middle ground 4:1 to 10:1.
Q3 might be from 5:1 to 12:1.
Q5 gives ranges from 7:1 to 20:1.

So either would be good choices, but a quantization that provides a range 4:1 to 10:1 would be a winner as it may be the best middle ground. We know with CinemaDNG 4:1 was a very popular choice as “size matters” when it comes to storage. Whatever the middle ground becomes, it will suit a lot of media bitrates on the cards we have and remove the curse of dropped frames.


Sent from my iPhone using Tapatalk

Andree, you may well be pleased with Q5 if that’s all you tested, but I’d recommend you test Q0 alongside every Q5 test and see if you notice a difference. I think you will, but you are the judge if the difference matters.


Sent from my iPhone using Tapatalk

Andree, you may well be pleased with Q5 if that’s all you tested, but I’d recommend you test Q0 alongside every Q5 test and see if you notice a difference.

As you can see in my post above, I tested all the flavours of braw in the static scene. I will do the same for a scene that includes motion and detail, like foliage or water.

If the static scene is any indication, I will opt for more compression more often than not. Especially if the intention is to shoot video (no still grabs) with delivery around 4K.

Regarding the many Q flavours, you also seem to agree with what I write higher up in the thread—I agree that your breakdown seems about right. I guess it becomes a question of how much one can accept as 'max compression'. It's hard to say since there is no white paper explaining the behaviour of the Q-modes.

Anecdotally: when I go through my sequence Q0, Q5, 3:1, 5:1, 8:1 and 12:1... sure there is a difference... but there is no motion blur and it's pixel peeping with a direct comparison.

And it's 100% clear to me that if I were to shoot a sequence in any of the formats and uploaded the end product, no one would EVER be able to say what it originated as.

The less uncompressed formats will attract the same kind of people who store their speaker wires in the freezer to "rest and realign the atoms" for a purer sound on their HiFi setup. =)

Ha! I didn’t know about the testing of wires; thanks for the tip!

I agree it’s next to impossible to detect these changes in motion pictures to the point where you say, “Hey that must be Q5!” But it may be that someone with expertise in testing perception of various people could run videos produced with different compression and assess people’s approval ratings. I’d assume statistically some clustering of higher ratings would go to the least compression to a point. Uncompressed films might have people saying, “I didn’t like seeing every pore, every follicle, it’s too distracting and who knew Taylor Swift was really that ugly?”


Sent from my iPhone using Tapatalk

I’d assume statistically some clustering of higher ratings would go to the least compression to a point.

But you'd almost certainly be wrong! The few times these kinds tests have been performed, the results tend to be incoherent -- largely because viewers don't have the ability to distinguish the differences, or make sense of them. Or they end up preferring the worse performance over the better.

Speaker wire is actually a good example. You can buy lamp cord for $.75 a foot or a 5 foot length of de-oxgenated multi-twisted 11 dimensional wonder wire for $1399. Asking the owner of the latter to do a blind test is a great way to get into a fist fight.

Like everyone else, I've looked at all flavors braw on static scenes. And sure, I can tell the difference -- sometimes, on fine detail, at a 6x zoom, one foot from the monitor.

It's plain crazy tormenting oneself over this stuff. It really doesn't matter, nobody can see it.

Yeah, not really, that's "blind" imagination. Try complex scenes like I described, but as statics. Without the exact scene frame on all test modes, the running water etc is just going play with real figures.

Speaker wire, yeah thanks for the tip. . I used to be great at audio quality, and detecting ultrasonic frequencies (and seeing the inter pixel grid in a good cinema screen set back 1/3 to 1/2 theatre back. But what you have to look for is the impression, the enjoyment, you subconsciously get from a picture. Your subconscious systems pick up way more than your normal system do. That is where joy, emotion and like happen. Popcorn imagery might look alright, is not as subtlety nuanced.

Actually, I over spec my speaker wire to get a purer sound, but what's an extra $30-$40 dollers in total. Maybe somebody with the multiple thousand dollar speaker wire can hear some difference is some top end setup. But, me I'll keep my around $70au, total, wire. Sounds much better than what came with the speakers, and I can use it on a 6x more powerful system apparently. Just barely fits in the clips too, with a lot of coaxing.

Wayne, that’s the point I meant to make when I said the pores and follicles were a distraction. I meant distracted from the enjoyment of the artful dream of a movie; I can see where John is correct they can’t tell which has less compression and might pick the film with the most compression.

Totally agree if that higher compression may enhance the viewers’ enjoyment. Starting to think I should just trust Q5 as many people have suggested. It’s not only good enough, it might actually be the most enjoyable viewing experience for narrative film and therefore the best choice between the two: Q0 or Q5.

Thanks, again, John.


Sent from my iPhone using Tapatalk

Well, let me clarify my personal opinion regardless if we're talking images or hifi:

The joy a person might get out of high fidelity material and/or equipment can be 100% real, EVEN if it would be impossible to measure any ability to distinguish A from B in said person.

The satisfaction of knowing or believing that things are in a certain way generates 100% real emotions. This applies to spirituality too, obviously.

Now, again to clarify, the IQ differences between compression ratios here are real. I see it when switching from one to the other. But I would have no chance—even looking at my own frames above!!—to say which image it was if you only showed me one at the time, with let's say a 2 minute pause in between. Just sayin'.

I have, and John P has, seen differences, but do they matter to the viewer who has nothing to compare them too but just knows whether or not they enjoyed the visuals? Andree and I might also feel greater personal enjoyment in producing the highest quality of deliverable we can with the tools at hand, I’d love to be able to shoot Q0 all the time, but will a viewer appreciate that? I have for years advocated here to “put your best foot forward” meaning make it as good as it can be. But with my reference to pores and follicles on faces, I’m starting to question my own advice. BRAW sort of forces this discussion because we no longer have lossless raw that is perfect for rendering pores and follicles.

Our perceptions of quality HiFi may be no different to a degree. Remember the Beatles rebuffed those who wanted them to record in stereo when they said they preferred delivering mono sound, incredible as that may be. It’s true, not an urban myth. Even so, if I have the choice between a thin wire and a thick wire, I’ll buy the thick wire.

I’m glad we are having this debate over detail. We are united in that we want the best colour if we can manage it and if 12bit BRAW does take off, I can imagine a day in the future where others may not shoot ProRes 10bit. If my BRAW shoot this weekend goes well, I may make the switch official from ProRes 422 HQ to BRAW on the BMPCC4K and from ProRes 444 to CinemaDNG on the URSA Mini 4.6K PL.


Sent from my iPhone using Tapatalk

I say, actual opinion and generalisation, that if you can pick it viewing a general image flipping back and forwards, in a good field of view, or up close if viewing picture levels, then your subconscious might be able to pick it in normal motion viewing. As you remember, I picked the cdng as more enjoyable due to contrast preservation on your building example (forgetting the massive detail sacrifice for the time being). I think braw is being overrun to keep file sizes down. Another thing it is doing is sacrificing detail so that there is less difference between Q0 and 12:1 on simpler scenes. Again, we need a cdng level Braw option, but maybe a little bit fixed up and not over sharpened.

During my analysis of hyper realism art decades ago, I could break down the mechanics of how the effect worked, which is a bit like what Braw does, but without the deathly lack of sharpness/detail. I suppose if we had 16k Braw cameras for 4k delivery, we would be reasonably OK, but what would the data rates be like? But Braw is part of the popcorn look, and you could just simply get a pro HDTV camera to get more of that (whoops) the earliest Panasonics. So, yes people enjoy popcorn, but if you want a more pro movie nuanced feel, something better is needed.

I wonder of any series production has been stung yet, updating their cameras to suddenly find they are stuck with only Braw, where they had previously been set up with look designed for cdng?

Now, there is something I'm going say. If you are going deliver at consumer level, better original recordings can still help your delivery quality. The best compression codec has only got what you feed it to attempt to preserve the image. If it's already degraded, than the compressor may degrade it further. If it wasn't Braw, we would be talking 3:1-5:1 level, depending on content. Braw doesn't appear to have a better compressor to warrant lower compression.

We have been advised that Q0 can be even less compression that the 2:1 indicated in the charts published in the manual or on the BMD webpages describing BRAW. Still it’s not lossless raw, but perhaps not far from it for images with lots of detail and contrast. I just don’t have enough experience working with it, but I’m using on the low light client shoot that is now Friday. No confetti and given the scene, will be closer to 4:1 or 5:1 compression I suspect than 2:1 with the predomine of low contrast shadows.


Sent from my iPhone using Tapatalk

2:1 should be nearly lossless, particularly here. But that's where it dips down, and it might only do it for very complicated frames. But is that twice the file size over Bayer size? If so (calculate frame sizes times bit depth times frames a frames per second and compare to file sizes, and if it's not over 3 times bigger, it's probably not).

Yes, less than 2:1 compression will only appear in very few circumstances of which they like to give the example of a shot of confetti that has a significant contrast with the rest of the frame. However confetti is in motion and might blur enough to not require edge detail. I think there are lots of other examples but experience will help us recognize when less than 2:1 (compared to uncompressed data) may be used. The thousands of shiny leaves in my backyard might do it.

But I don’t see that happening with average interior shots in a narrative film without intentionally complex props like a hanging lace curtain and so on. I think my media can handle anything the BMPCC4K camera can throw at it shooting 24p so I’m not worried now. Looking forward to pushing the detail in scenes which contributes to the beauty of the shot while knowing the talent will be treated kindly.

The nice thing about BRAW is that these detail decisions are dynamic and very local to the things that require more detail, whereas the human face will likely have less detail due to the way it’s usually lit.


Sent from my iPhone using Tapatalk

I would not be sure about that. A good lossy codec should identify important things and devote more data to them. That was my suggested solution around 15 years ago, when I shared my image preservation preference list, and compression quality then leaped. Older camcorders orders were also set up for different skin tones depending on region. I suppose to preserve those better and devote more data to them.

But I'll say this. If you devote more to detail on the axis and placement of your standard framing, preferring skin face detail and human shapes, then central object, then horizontal axis, then vertical, with more on objects and positional placement regions (remembering, seeing a face or human will override). But also order things by what's in focus. You can subtly adjust detail and compression across the scene based on what people look at, plus my previous list. The blend from one quality to another from region to region may not be noticeable, as people concentrate on the things they look at. Of course, you add the complexity preservation factor in these different regions customised, and at the end (but with a minimum). So you can mix 12:1-2:1.

Wayne, it’s a given that Q0 will concentrate on areas in focus as that’s where you will have greater contrast differences in edges. I doubt it’s concentrating on items related to facial recognition as we are talking about preserving detail that is there throughout the frame. Eye lashes will get the full treatment but not cheeks. Enhancing detail but hopefully not creating false detail on faces. If contrast and edge detection tend to be areas that receive the greatest fidelity, an actor’s face is not likely to drive the bitrates to a maximum.


Sent from my iPhone using Tapatalk

Especially not if you use a narrow DoF, the blurred background will save a lot of bandwidth.
I have seen the highest bit rates in Q0 with a very sharp lens (Zeiss Contax), stopped down to f8 and filling the frame with leaves against the sky.

I would be careful with doing different things in the face differently (though hair is difficult) it can become cartoon like. I'm talking about doing whole areas or objects with equal encoding across the object, and blend between areas. When filming we use these things to draw the audience's attention to them, as the audience is drawn to them as well. What they are watching should get more quality, and data saved elsewhere.

These are visual perception things. Sometimes audiences are searching the background or around a scene, so that's an exception. This is how you expand visually lossless (but gets to be attention visually lossless). It is just good lossy codec behaviour.

Wouldn't it make sense to introduce a Q3 mode as another option between Q0 and Q5?

Yes, please!

In my experience, they already overlap.
If you have very low detail in Q0 vs very high detail in Q5, the data rate is about the same.

In my experience, they already overlap.
If you have very low detail in Q0 vs very high detail in Q5, the data rate is about the same.

Not according to the official rates Q0 lowest for 4K is 82 and highest for Q5 is 58

So I think people are hoping for something like this