60Mbp/s vs. 100Mbps - Supprising Video Samples

[Cliff Totten]

OK folks, I have been all over the internet shouting at Sony to give the PXW-X70 a 100Mbp/s bit rate option.....

So, this weekend, I decided to do an AX100 codec test just to see what the real difference truly is. I shot the same scenes with the same exposure and just toggled files with the two different bit rates. I then did a split screen and made identical zoom and crops on both shots. These all have a 400% or more zoom so that we can really "pixel peep" these things so YouTube's crappy resolution doesn't get in the way.

Here is the test result:

YouTube (Short URL)

I must say that I'm blown away at how similar these shots tested! On some, the 100Mbp/s was SLIGHTLY less blocky while on others, the 60Mbp/s appeared to have more noise reduction? The bottom line is that they don't look far away from each other at all...and this is REALLY strange.

Is Sony "padding" their 60Mbp/s codec to artificially "inflate" it to 100Mbp/s?

Is the Sony AX100 codec simply not able to "efficiently" encode 100Mbp/s to it's full potential?

In UHD, adding 40 megabits to 60 should encode ALLOT more bits deeper into the blacks. I realize that the mid-tones might not see allot more efficiency but the shadows SHOULD be getting those extra bits allocated to them. But here, this doesn't seem to be the case. Why?

I'd love to see similar tests form you fines folks out there. If you have an AX100, please take shots with both codec bitrates, combine them with a deep zoom-in and let's all take a look.

This is a very telling test result so far....very strange.

Your thoughts?

[John Nantz]

Cliff,

Dang, this is good. I like a good mystery story and this looks like one.
.

Originally Posted by Cliff Totten

This is a very telling test result so far....very strange.

And, those are some very good questions.

Because the pictures look the same, I assume both cams were on a tripod so we could rule out any input from Image Stabilization.

Well done. "The proof is in the pudding" as they say.

[Andy Smith]

Yes, I absolutely agree (I did raise a post before asking if anyone could see any difference). The only difference is that 100Mbps produces much bigger files. Another question would be does the 100Mbps stand up any better to colour grading than 60?

[Cliff Totten]

There is one very telling test that I have not done yet. That's 4K 60Mbp/s internal XAVC-S vs. 4K ProRes from the HDMI port.

By all rights, when compared side by side, the ProRes should look significantly better than 60Mbp/s.

If the two look exactly the exactly same? Wow...that could mean only one thing: The image quality "bottle neck" exits BEFORE the internal codec and BEFORE the HDMI port. In other words, the image quality is "limited" or "reduced" before it gets to the codec or HDMI port and there is nothing that can be done to improve it. (The image quality would be the same at 60Mbp/s, 100Mbp/s or ProRes)

Again....only further testing can prove this. But...this will be an excellent test to perform and could reveal some big internal processing AX100 "secrets".

Lets see....I don't want to jump to conclusions here but that will be a very important test.

EDIT - New info - Just did a side by side with internal XAVC-S 60MBp/s vs. 4K ProRes HG off the AX100's HDMI port. Good NEWS!...the ProRes looks significantly sharper in the shadows. It definitely has a very different noise structure and DOES hold detail in the shadows better. NOTE...Sony employs noise reduction into it's XAVC-S codec processing. The HDMI port does not have the noise reduction. As you know, Noise reduction softens the image. This is another reason why ProRes will look sharper. If you are using anything above 0db, with HDMI, you will need to add noise reduction.

Whew! I think I can put that concern to rest. The AX100's HDMI does seem to be pre-internal codec.

Only question now is: What is wrong with the AX100's XAVC-S codec? Why is 60Mbp/s and 100Mbp/s 99.9% identical???

[Mike Buckhout]

I should think we also need to do a side-by-side of the AX100 60Mbps 4K, and the X70 60Mbps 4K.

[Jos Svendsen]

Hi There,

It is not that surprising as the number of bit per second is enabling you to handle changes in the picture better, as the numbers goes up. If you have a totally static picture you would not be able to see any difference between a recording with 1 Megabit/S and a recording with 100 Megabit/S. Another thing is that AX100 and X70 is not running the same subversion of XAVC. X70 is using XAVC HD1080 Long GOP, and will probably use XAVC QFHD Long GOP in 4K version. AX100 is using XAVC S HD and XAVC S QFHD. This means that XAVC QFHD Long GOP at 60 Megabit/S might be equivalent to XAVC S QFHD at 100 Megabit/S. Both variants are variable bit rate, and the test recordings might not have demanded the full bandwidth available. XAVC is a very effective CODEC.Try to record waves at sea, to ensure most pixels changes between every frame.

The best
JOSS

[Cliff Totten]

Well? "XAVC-L" and "XAVC-S" are really not "codecs". They are really just industry standard MPEG h.264 codecs wrapped in .mp4 and .mxf containers. They follow the MPEG h.264 specification completely. There is nothing unique or anything that exceeds the h.264 spec.

Sony has created a file and folder structure and built an XAVC metadata structure to hold all kinds of info but this doesn't affect the h.264 codec quality. This is really what "XAVC" is.

Sony does have some neat processing tricks for "look ahead" VBR encoding but again the files are just pure h.264 as defined by MPEG long, long ago.

I'm going to do some side by side tests that contain allot more motion next weekend. Let's see how that looks. There is no reason why 60 and 100 Mbp/s should look so similar with high motion scenes....unless Sony "wants" it that way.

[David Heath]

Originally Posted by Jos Svendsen

Hi There,

It is not that surprising as the number of bit per second is enabling you to handle changes in the picture better, as the numbers goes up. If you have a totally static picture you would not be able to see any difference between a recording with 1 Megabit/S and a recording with 100 Megabit/S.

Not necessarily true.

It depends how the inter-frame codec is implemented. What you say may have been true for earlier implementations, but nowadays such as XDCAM HD uses dynamic allocation of bitrate between the I-frames and difference frames - earlier inter-frame codecs tended to be fixed.

In simple terms it means that for static scenes the I frames get allocated a lot of bandwidth, difference frames not very much. With a lot of movement a far higher percentage of datarate gets allocated to the difference frames - at the expense of allocation to I-frames.

Assuming 50Mbs, and 50fps and a GOP of 50, then in an extreme example imagine the image is totally static - our ideal encoder may allocate to 50Mb to the I-frame and none to the difference frames. In the other extreme (every frame with a different image) it may allocate as much to every difference frame as to the I-frames - so 1Mb to every frame. Effectively it becomes an I-frame only codec!

Practically, the differences won't be anything like as extreme, both in terms of content or variation in allocation, but hopefully it helps to give an understanding of why the results will be so much better than a simpler encoder which always allocates x% to the I-frames and y% to difference frames.

And this is one reason why XDCAM gives so much better results than HDV, in spite of both being fundamentally MPEG2, and XDCAM EX not having that much higher a bitrate overall than HDV. Beware of drawing conclusions solely from headline statistics like bitrate.

[Dave Blackhurst]

There may well be a point of diminishing returns that only will be noticeable under certain shooting conditions. I'd expect that high motion would benefit from the higher bitrate, a "still motion" scene, not so much.

Not sure what you're monitoring the footage on, but even that may be a limitation as well - I've come to suspect that there are a few "glitches" where my cheap-o "monitor" hits the limits in color handling - nothing major, but certain colors in the red/hot pink spectrum seem to glitch a bit. Detail wise, there are many things that you won't be able to see on a 1080 screen no matter how much you zoom in, but will be obvious on a 4K screen.

There are many points along the chain where things can be degraded. I think this is why when I tried direct HDMI in to my TV/monitor, the onscreen image displayed from the RX10 (which is only 1080p) looked significantly better than the output from the "4K" AX33 I was evaluating... the starting point of the lens/sensor was obviously several notches "better" on the RX10 (as was the AX100).

[David Heath]

Originally Posted by Cliff Totten

Sony has created a file and folder structure and built an XAVC metadata structure to hold all kinds of info but this doesn't affect the h.264 codec quality. This is really what "XAVC" is.

Not true - such as H264 really only uniquely defines how a file is to be DECODED. It doesn't precisely define how it should be encoded as long as the result is able to be decoded according to the standard.

Hence manufacturers have quite a lot of flexibility in coder design, and are able to take advantage of new techniques as they develop - with XAVC Sony are implementing H264 at level 5.2. Which will give somewhat better quality than earlier implementations of H264 coding at the same bitrate and assuming all else equal.

[Cliff Totten]

Agreed David,

Yes, any company can use as many ENcoding tricks that they want. I have read that Sony has some key patents on their ENcoding techniques including tricks they have with VBR specifically.

Sony has full access to the existing MPEG h.264 toolset, like CALVC and CABAC and loop filters and blocking tools.The "look ahead" schemes and encoding is particularly powerful.

H.264 "profiles" like "baseline", "main", "high" and more contain higher and higher math and tool sets. Each level brings more tools galore! there are tons of tool sets. 8x8, 4x4 adaptive transform, slicing tools, quanitzation scaling, prediction techniques...blah blah...haha, it's allot of calculus and geometry!

"Levels" determine simple limitations for frame size, bitrate, frame rate parameters.

This is all wonderful but Sony can't put in new calculations that the h.264 specification can't understand. If Sony did that, than no h.264 complaint DEcoder would be able to make sense of it. One tiny "extra" or "unknown" calculation could literally "break" the entire file from playing back at all! Looking at H.265 HEVC,...this codec contains all the tools that H.264 has. In fact, h.264 is at the core foundation of HEVC! MPEG simply took h.264 and threw a TON of new math on top of it and raised the level restrictions way higher. No h.264 DEcoder can play any h.265 file because there is an enormous amount of "new" calculations that that poor h.264 DEcoder just doesn't understand.

Sony can do whatever they want on the ENcode but it must not "break" any of the h.264 calculations used for a fully complaint DEcode. So, the end result of all of Sony's XAVC efforts has an end result or "limitation" of being stuck on being 100% h.264 complaint.

But yeah,...ENcode to your hearts content but it's still gotta fit into the H.264 white paper limits on DEcode.

[Zenes Petrusin]

Detailed information about XAVC is on sony web page.

XAVC White Paper : United Kingdom : Sony Professional

[Zenes Petrusin]

Hello,

my X70 has license for 4k, this is short samples doing fast panning. Original MXF files is for download here. Shutter speed is 1/50 in 25fps

Sony PXW-X70 4K 25p 1/50 #1 - YouTube

Sony PXW-X70 4K 25p 1/50 #2 - YouTube

Sony PXW-X70 4K 25p 1/50 #3 - YouTube

Original MXF 4k from PXW-X70
https://ssl.unlp.sk/x70/Clip0045.MXF
https://ssl.unlp.sk/x70/Clip0047.MXF
https://ssl.unlp.sk/x70/Clip0052.MXF

[Piotr Wozniacki]

Thanks Zenes; unfortunately in Vegas Pro 11 the clips play at some 5 fps at Best/Full (watchable from Preview/Quarter and down). What the heck - this is Sony's own codec! Well, I guess proxies are the only way to edit - even a single track...

Something is terribly f***ed up in Vegas, because at Best/Full, CPU is only loaded up to 50% and GPU - up to 25%.

[Zenes Petrusin]

try use vlc, play this 4k smoothly on 4cores i5