Other: XAVC-I vs. ProResHQ

[AmbiSonde]

This is an FS7 XAVC-I 4k image at 100%. Macroblocks are smaller but visible and notice the horizontal stripe compression artefact.

XAVC-I and SLog3 on the FS7 was extremely prone this artefacting. The compressor would favour areas of high detail and motion and anywhere which had flat low detail would be compressed heavily. This would mean skies were particularly badly affected. If you just transformed SLog3 to REC709 you would not notice any compression. However, once you applied saturation to the very pale pastel blue skies common with SLog3 the compression damage was easy to expose.

Any nonlinear tone or colour adjustment could cause the image to shimmer with noise. Subtractive saturation adjustments would be impossible, colour warping a very common grading operation can quickly separate the compression posterisation.

 

[AmbiSonde]

XAVC-I with SLog2 faired much better. The Tetris style compression artefacts are gone, there were no obvious horizontal streak artefacts just fairly typical grainy macroblocks.

SLog2 held up to grading much better as a result and even the image was pushed hard the macroblock noise had more of a grain appearance which did not catch the eye so much.

 

[AmbiSonde]

This is from the FX30 XAVC-SI. I had to apply absurd levels of contrast to get any macroblocks to show in the sky. This has the same bitrate as all of the previous FS7 examples but shows far less compression artefacts and those there are are much finer even than the FS7 SLog2 example.

Sony has obviously vastly improved XAVC-I since the previous generation. As I suspected.

 

[AmbiSonde]

As I have previously stated SLog3 is noisier than HLG3 and S-Cinetone but that noise shows up from middle grey down. I have yet to see any artefacting in FX30 footage anything like the FS7 even the XAVC-S behaves well.

So Doug, you see there were very good reasons why FS camera users would shoot SLog3 with ProRes and producers would request it because ProRes did not produce Tetris style macroblocking. There was some banding to deal with due to the 12 bit raw output of FS cameras but no horrible artefacts lurking under the surface to catch out professional colourists who do expect to push the image around.

Sony has clearly updated XAVC-I so it no longer trashes areas of flat colour which is a major plus. I have several more examples from my FX30 that show how improved XAVC-I is but I don't think I need to labour the point.

Anyone not spotting such awful artefacts can only be described as an amateurish clown.

 

[Bassman2003]

Good for you to post the examples. I do not know why it is so difficult to get to this stage? Maybe we can all learn something.

BTW, you were doing so well until the final sentence... sigh.

I do not doubt that there would be constant development with all codecs since the microprocessors get faster and faster to allow for more processing. Whatever is going on with the FS7 first clip is crazy and seems like far more than macro blocking. Did you show this or a similar image to Sony and they brushed it off?

 

[NorBro]

There's a lot of energy going into all of this, lol.

We've been talking about sky banding and chasing better quality formats since the DVX100 - or maybe more the HVX200 - while some people on YouTube are making hundreds of thousands of dollars shooting pixel-binned & line-skipped 1080p.

There's no point in arguing about it because who cares? You know?

For many, this is moot as they will rarely ever have skies in their frame, or are not worried about any other minimal artifacts from internal recording (the point being that if the macroblocking or other imperfections were major enough then the cameras would be recalled or have FW updates).

Film was messy and imperfect for like 100 years. [Dust, scratches, grain, tears.]

For others, it could be absolutely highly-encouraged to explore all and any options if even a single case study for that user concludes ProRes is better than XAVC-anything for any given situation in their lives.

 

[AmbiSonde]

Good for you to post the examples. I do not know why it is so difficult to get to this stage? Maybe we can all learn something.

BTW, you were doing so well until the final sentence... sigh.

I do not doubt that there would be constant development with all codecs since the microprocessors get faster and faster to allow for more processing. Whatever is going on with the FS7 first clip is crazy and seems like far more than macro blocking. Did you show this or a similar image to Sony and they brushed it off?

I sent images and my comments on how to repeat but all I got back was there's nothing wrong with my camera.

The FS7 could shoot images without these artefacts, if I shot a clip of the sea with every single pixel on-screen moving you wouldn't see these artefacts but shoot the sea and have a third of the screen devoted to a blue sky then you'd get the Tetris blocks in the sky (if you applied sat + grad in post). I tried shooting skies with a grad filter and/or a circular polariser just to get some pre-compression saturation and contrast into the skies but I ended up just shooting SLog2, it was far more robust even in the blacks and I knew it would never artefact like SLog3 and it simplified my shooting gear.

I think I understand why the Tetris blocks appear, it's due to the very flat curve of SLog3 making pixel values so close that the compressor makes bigger macroblocks, it's due to SLog3 sample distribution which can be poor in low dynamic range scenes and the fact the compressor is tuned far too much towards motion and detail preservation. When all these things coincide the compression goes a bit mad.

It was my experience with the FS7 and SLog3 that made me compare SLog3 and HLG3 on the FX30 and I found that HLG3 had even smaller macroblocks and much less prone to that rainbow noise in the blacks. HLG3 is a close relative of SLog2 just a tad more contrasty, it grades impeccably well and I cannot recommend it highly enough. To my surprise S-Cinetone is also pretty good and my first choice for low dynamic range scenes, super clean and makes full use of the 10 bits with a big fat waveform.

Anyway, when I find the time I will post HLG3, S-Cinetone and SLog3 comparisons images and breakdowns but I'm very busy so I don't know when I can get it done. It's really only going to be of interest to those who do deeply care about the image they're needing the most robust file for grading. SLog3 + XAVC-I is much better this time around but not as good as HLG3 + XAVC-I.

 

[AmbiSonde]

For many, this is moot as they will rarely ever have skies in their frame, or are not worried about any other minimal artifacts from internal recording (the point being that if the macroblocking or other imperfections were major enough then the cameras would be recalled or have FW updates).

It's not just skies. Shoot an interview with defocussed green grass behind the subject and that flat colour will be heavily compressed. In post you want to add a bit of visual interest to that grass so you give it a bit of a colour warp from yellow to blue (a common film like grade) you'll soon see those macroblocks appear.

The FS7 was used most as a rec709 camera for broadcast so I doubt many of the broadcast users ever used SLog3.

Well, the macroblocking on the FS7 was a major imperfection but there was no recall. Sony did fix the compressor in the FX series so they must've known about it.

 

[NorBro]

I've owned and used over 100 cameras so I genuinely do not doubt anything you're saying or seeing because I've been through it all with these systems, but on the other end being that the FS7 was one of the world's most successful cameras of all time I would also have to think the majority didn't see what you do/did (and most people I knew only shot in Log).

 

[Doug Jensen]

Imagine not noticing macroblock artefacts like this.
FS7, XAVC-I SLog3 HD crop at 100%

Thanks for posting some examples, even if they are just stills. If you don't mind, a few questions:

1) What version of Resolve are you using?
2) What LUT are you applying in post?
3) Will you export your grade and make it available for download so that we can see exactly what is being done to the clip?
4) Will you post the original clip for download so that others can attempt to recreate the phenomenon?
5) Where is your XAVC-I vs. ProRes side-by-side comparison of the exact same shot recorded simultaneously from the FS7?

Thanks in advance.

 

[cyvideo]

This was an interesting video and I like his information.

Agreed. I also wish he was a Resolve user, Notwithstanding that he does make and does demonstrate and dispel some of the myths around various codecs and compression algorithms. I think he is aware enough to realise that Premiere has its limits in the RAW workflow, hence his doing the same tests with the Canon RAW tool with considerably better results. I felt it was one of the better video examples of the differences in codec results I have seen on social media. Agreed with the comment that there are no appreciable differences in any of the RAW workflows.

Chris Young

 

[cyvideo]

Sony has clearly updated XAVC-I so it no longer trashes areas of flat colour which is a major plus. I have several more examples from my FX30 that show how improved XAVC-I is but I don't think I need to labour the point.

The biggest problem I see in your examples is that they are very symptomatic of 10-bit video being processed through an 8-bit path and then encoded back to 10-bit. Examples of this have been well and truly demonstrated on this forum. Admittedly some years back, but examples like you are showing when processed through a 100% 10-bit pipeline don't exhibit the issues that you are stating are a product of a poor quality codec.

Re Sony's XAVC codec. Not improved or changed. Sony may well have improved their hardware processing, but Sony's XAVC codec along with Panasonic's and Canon's iteration of H.264 at up to High 4:2:2 Profile Level 5.2 are registered and ratified by SMPTE. These are broadcast standards that broadcast networks around the world can then use to budget for technical developments in their own organisations. The whole deal with SMPTE ratification is that organisations can abide by these standards and know there are no nasty surprises coming in the change of codecs that is going to impact the technical and financial investments they have made. It also lays out a clear path for software and hardware decoder developments. It's pointless changing a codecs structure if there are no decoders out there to handle those changes. Sticking to these standards also guarantees that a fully operable set of codec interchange specifications exist for the interchange of material between broadcasters and production houses around the world. A codec standard once ratified by SMPTE cannot be changed without the company re-submitting an application with the proposed changes. As long as those changes don't impact the existing ratified specification that organisations have signed to then it will be accepted and all signatories to the ratification will be notified. The last change to Sony's H.264 XAVC codec was back in 2019 when Sony introduced their new XAVC QFHD Long 422 200 codec. Which fully complied with all the ratified decoder classifications that were part of the existing original ratification. So in short, the answer is NO. Sony, along with Panasonic and Canon, none of them have "changed" their AVC codecs, in either 8-bit or 10-bit variants. They have added new variants but have not changed the specifications.

Without the guarantee of locked in ratified specs, the following companies and organisations would not be investing millions is developments.

The original signatories to the broadcast AVC H.264 standard:

Abekas-Adobe-AFrame-Apple-Calibrated Software-Cinergy-Astro-Autodesk-Avid-Blackmagic Design-MaghWare-Codex-Colorfront-Corel-Cyberlink-Deluxe Digital Studios-Digital Vision-EVS-Fairlight-Filmlight-Firefly-ForA-Grass Valley-Harris Broadcast-Hitachi-Ibex-Imagination-Matrox-MTI Film-Nablet-NEC-NLT-Panasonic-Pegasys-Pixel Power-Quantel-Rohde & Schwarz-Main Concept-Sakura-Eiki-Sobey-Sony Media-Sony PCL-Tata-Telestream-UP4-Vanguard Video-Magix Vegas Pro-Village Island-YoYotta-Zaxel.

Chris Young

Sony's last H.264 XAVC iteration that was approved by SMPTE:

https://pro.sony/s3/2019/04/3016172...point-for-advanced-media-architectures-EN.pdf

 

[cyvideo]

3) Will you export your grade and make it available for download so that we can see exactly what is being done to the clip?
4) Will you post the original clip for download so that others can attempt to recreate the phenomenon?

Yes, as Doug suggests. Please do supply us an original FS7 sky clip and let some of us have a go at it for you. You may be surprised at the results. The proof of the pudding is in the eating of it. At the moment, you are only offering us a sniff of it.

Chris Young

 

[AmbiSonde]

I've owned and used over 100 cameras so I genuinely do not doubt anything you're saying or seeing because I've been through it all with these systems, but on the other end being that the FS7 was one of the world's most successful cameras of all time I would also have to think the majority didn't see what you do/did (and most people I knew only shot in Log).

I've owned most of Sony's cameras since the PD-150 and used nearly all the others that I haven't personally owned.

People shooting Slog3 would've first noticed an increase in image noise when grading, you see the horizontal stripe artefacts first in my experience. I'm sure most people would have simply backed off the grade and lived with a REC709 look. Going from SLog3 to REC709 was fine but that was pretty much all the camera would allow because these underlying macroblocks turned a 10 bit image into an effective 8 bit image. I always wondered if these artefacts were Sony's way of limiting the performance of the FS7 so it didn't step on the toes of more expensive models? There was little competition to the FS7 at the time but Sony no longer have that luxury as there are many more camera makers vying for the market. If it wasn't an imposed limitation maybe it was a hardware limitation or caused by the 12 bit internal processing but that doesn't explain why SLog2 was so much better.

Images from my FX30 grade superbly well, I have very little limitation where I want to push the image, same bitrate, bit depth only the compression scheme is significantly better. I am so blown away by this tiny camera and the image it can create. Which brings me back to how this whole situation started. DVX used to be a place to share knowledge about cameras, I had the temerity to share my knowledge that on balance HLG3 provided the most robust high dynamic range image for grading, someone took umbrage at that so here we are.

 

[AmbiSonde]

Thanks for posting some examples, even if they are just stills. If you don't mind, a few questions:

1) What version of Resolve are you using?
2) What LUT are you applying in post?
3) Will you export your grade and make it available for download so that we can see exactly what is being done to the clip?
4) Will you post the original clip for download so that others can attempt to recreate the phenomenon?
5) Where is your XAVC-I vs. ProRes side-by-side comparison of the exact same shot recorded simultaneously from the FS7?

Thanks in advance.

The version of Resolve has nothing to do with it.
There are no LUTs
no
no. You have an FS7 you can easily recreate the problem for yourself.
I have looked through my extensive clip archive and I cannot find any ProRes footage. I think I must've deleted it all when I moved to PC. I'm disappointed because it would've proven the next point which you took umbrage with that Sony's FS7 era XAVC adds sharpening and the ProRes recorder we were using didn't producing a more aesthetic image.

Despite what CYVideo says about XAVC not changing, XAVC-I in the FX series cameras is a very different beast to the XAVC-I in the FS7. I can see very little need for the vast majority of people to need ProRes recording with the FX cameras as the native codec is now so good. In fact XAVC-I, HS and S are all better than FS7's XAVC-I output.

However, having the ability to record 4k ProRes in 12 bits from the RAW output of the FX series should not be underestimated. If you need it, you need it. One day I hope Sony will support 12 bit in XAVC as XAVC does support 12 bits.

 

[AmbiSonde]

The biggest problem I see in your examples is that they are very symptomatic of 10-bit video being processed through an 8-bit path and then encoded back to 10-bit. Examples of this have been well and truly demonstrated on this forum. Admittedly some years back, but examples like you are showing when processed through a 100% 10-bit pipeline don't exhibit the issues that you are stating are a product of a poor quality codec.

Are you suggesting Sony crunches SLog3 to 8 bits then encodes it as 10 bits?

For the avoidance of doubt, I'm using Resolve correctly, probably more correctly than the vast majority of people here given the bad advice I've frequently read.

What does SMPTE actually cover though, frame rates and bitrates or right down to the bit level output of the encoder? Please provide links so I can read because the FX30 XAVC-SI output looks completely different to FS7 XAVC-I. What is SMPTE verifying?

 

[AmbiSonde]

Yes, as Doug suggests. Please do supply us an original FS7 sky clip and let some of us have a go at it for you. You may be surprised at the results. The proof of the pudding is in the eating of it. At the moment, you are only offering us a sniff of it.

Chris Young

I won't be surprised at the results although I am endlessly surprised at the arrogance on display in this forum. I am completely confident in my ability to colour manage images.

Dougie has an FS7 I'm sure he's firing it up right now in an effort to disprove me. That's how it should be but I have countless other examples I could post if I trawled my old footage archive. I could even fire up the FS7 again if I had the time but I don't. I could do a forensic examination of A vs B if I had all the time in the world but I don't. Believe me or don't it's up to you.

Here's a recipe to recreate the issue, Shoot some trees blowing in the wind, devote about 2/3rds of the screen to the trees and the final 1/3 to a flat blue sky/clouds you'll see the Tetris macroblocking once you apply contrast to the sky. Add gain and reduce gamma to split the posterisation levels to make them more obvious.

However, if you shoot just a clear blue sky there will be no Tetris macroblocking. There seems to be a threshold of detail and motion that the XAVC-I encoder hits and it doesn't leave enough band width for flat motionless areas. If every pixel on screen is moving e.g. a shot of a wavy sea you will not see tetris macroblocks either.

Despite what you say about XAVC not changing it is on version 2.1 now and try as I might I cannot force any ugly macroblock artefacts out of the FX30. The worst compression artefacts are when using SLog3 but it's a million miles away from the FS7. All other tested gammas the artefacts could be described as a film grain like and you have to add truly absurd levels of contrast to break the image. I'm blown away by the quality of the image and the quality of the compression out of the FX30.

 

[cyvideo]

I won't be surprised at the results although I am endlessly surprised at the arrogance on display in this forum. I am completely confident in my ability to colour manage images.

I'll tell you now that if you were to come to me having posted a job which contained these kinds of visual artefacts, I'm pretty sure I wouldn't be confident in your ability to manage a color workflow. What you would be telling me is that you haven't been able to solve your blocking issue. But that would only be my opinion, having edited many programs for broadcast, like fifty a year for many years I think my opinion may hold some validity. I don't know why I'm bothering, as you obviously aren't keen to deliver up one of your XAVC S-LOG3 sky sample clips for us on this forum to have a look at. But I will offer an opinion. Whether it's of any help to you or not, I don't know... but it may help some others who have seen this problem and are open to workflow suggestions.

Who knows, the FS7 sample you display could possibly be from a problematic camera. Though I'm 99% sure that is not the problem with the sample you uploaded, as I've seen this issue before. As you haven't fronted up with a clip, I'll take the liberty to deliver an example of XAVC-I robustness.

This following example is an FS7 S-LOG 3 S Gamut 3.Cine XAVC-I sky and tree shot, to compare apples with apples, a little like one of your examples. This example has had an insane, and I mean an insane, grade thrown at it. The peak level exposure on the clip as exposed was just on 60 IRE, 61 being Sony's recommended peak level. I applied the following to create the insane grade. As listed on the attached pix. 1) a Color Transform 2) Custom Curves (for level adjustment) 3) a 100% Saturation Boost, yes 100% and 4) added to that a 100% Color Boost, yes another 100% then 5) then to top off the madness I added a x 2 increase in Hue vs Sat Curves just in the blue channel. A totally insane grade, as can be totally verified by the accompanying screenshot of the scopes. The JPG you see here was composited from the TIFF image grabs off the Resolve timeline. There is no indication of any disastrous image brick wall blocking breakup as you have displayed with your images.

In the example I have supplied, all of it was processed though a full 10-bit pipeline. A 10-bit camera file processed on a 10-bit timeline, with a 10-bit grade through a 32-bit desktop color path at 10- bpc color depth. This, of course, outside of broadcast hardware can only be delivered on a PC via a 10-bit graphics card to a 10-bit monitor through a 10-bit Display Port. So if the grabs you have presented on the forum are have been grabbed from an HDMI source I can understand why you are getting the results you are. Sure, I can see encoding artefacts on my TIFF file grab of the XAVC-I file. But then I would also expect to hear total distortion on an audio recording made with the levels jammed against zero with max clipping. Because that's what this grade is, a totally destructive grade of an XAVC-I image. But it's still way better than what you have presented. Blame the codec last. Examine your workflow. If you want further evidence of the robustness of the XAVC-I codec upload one of your "terrible codec" files and let me attempt a grade on it. If I can't deliver a half decent looking grade out of it, I will be the first person to admit to it. BTW, I'm happy to supply full res images of any samples I post.

Chris Young

 

[cyvideo]

Are you suggesting Sony crunches SLog3 to 8 bits then encodes it as 10 bits?

No. What I'm suggesting is that you have some concatenation issues. 10-bit to 8-bit and then back to 10-bit in your workflow.

https://tech.ebu.ch/docs/techreview/trev_271-dalton.pdf

S.M.P.T.E. (Society of Motion Picture and Television Engineers) The organisation that sets video standards. Of which I was a member of for over forty years. They are the ones who ratify codecs for use in the film and television industry. Look under Standards.

https://www.smpte.org/

Chris Young

 

[Doug Jensen]

Excellent post, Chris. Spot on. You are 100% correct.
This is like someone telling us they have created cold fusion from sea water, or can turn lead into gold, but then refuses to reveal any details of how they did it so that ofher people can experiment and confirm or invalidate the results.