Other: XAVC Long GOP 4K 10Bit 4:2:2 200Mbps ANNOUNCED
- Thread starter alohype
- Start date
Cary Knoop
Veteran
XAVC is basically H.264, all those extra "features" seem to me mostly marketing.
Tom - I believe the 16bit from the F55 would & should be superior. I wasn't clear. I was speaking specifically about the raw coming from the extension unit on the FS7 - which is (I believe) 12bit. I only rented the back with Ody 7+ once to test it out - in the early days....and couldn't justify the added file sizes / workflow for my type of work. Others on this forum also proved to me that shooting raw with the FS7 wasn't justifiable.
cyvideo
Veteran
No Tom. That is what a lot of people don't get. 10-bit processed correctly to 8-bit can yield decent looking 8-bit graduations. If you start your 10-bit in the 8-bit domain then 8-bit it will always remain. There is no correct interpolation taking place from 10 to 8-bit when it is just dropped on the timeline. If you start by just placing your 10-bit footage on an 8-bit path the first thing you get is concatenation. The 1023 samples of the 10-bit depth is immediately broken down to into 255 levels. The NLE at that point is not automatically analyzing the image to decide whether it's 8-bit or 10-bit. You have told it to start an 8-bit process.
Let say part of you image has a fine graduation like a blue sky. Let's take one portion of this 10-bit image blue sky. It could represent four shades of blue out of those 1023 samples. When this is dropped into a 255 level process the first thing that happens is the 8-bit process looks at the image and decides because of its limited range it will put those four shades of blue into the closet single shade of blue it can match them to in the 8-bit domain. This happens for the next four shades of blue and so on, or to any other color where subtle graduations occur. Subsequently you will see banding and those single color bands are the closest approximation of the four shades that got lumped together. Most modern graphics algorithms can do a fairly decent 10-bit interpolation of four shades into one. But if you tell the graphics engine from the get go that this 10-bit image is an 8-bit image it will just do a crude division by four. Take basically any decent still camera that records its image to 8-bit JPEGs and go outside a shoot a scene with a decent amount of sky. In almost 100% of cases what do you get. A decent looking graduated blue in the sky from an 8-bit JPG on an 8-bit monitor.
In a still camera all the intelligent interpolation math is done by algorithms in the camera's hardware and firmware to deliver that smooth end result 8-bit JPG. With video processing we have to go through a different process. A process of taking that data and carefully processing it. With higher order video data and intelligent algorithms we can deliver a result that is similar to the 8-bit end result we see from the still camera. Putting it simply the correct process has to be applied to achieve that decent looking 8-bit graduation from your original 10-bit image. Below are two still examples grabbed from Breaking Bad. The file these stills came from is an 8-bit 4:2:0 file. As you can see both show no gradation breakdown in the very dominant blue sky because the original editing and conversions and color interpolation were done correctly. In this case the original acquisition was film using. Arricam LT, Arricam ST and Arriflex 435 and the negative formats 35 mm (Kodak Vision2 50D 5201, Vision2 200T 5217, Vision2 500T 5260). In other words the processing path was a long one through to the Digital Intermediates but was done correctly.
On a side note I was involved on the technical side of supplying 25 graphics systems to the ABC network here for their regional stations. All Australian capitol cities were running fully configured 10-bit Qantel Paintbox systems. They didn't have the budget for Qantel for all the regionals. What they ended up with was UK built G2 graphics system which were all 8-bit. Immediately the ABC ran into problems with file interchange between the Qantel and G2 systems because of bad 8-bit/10-bit concatenation issues on graduated backgrounds on their graphics packages. We ended up having to get Qantel and G2 to write a software utility to be able to get a smooth translation between the 8-bit and 10-bit graphic worlds to render out smooth 8-bit graduations. In other words the correct color space transform process.
Chris Young
Edit: More info on concatenation:
https://www.fujitsu.com/global/documents/about/resources/awards/ip-9500-concatenation_wp.pdf
Last edited:
cyvideo
Veteran
Yes Gary XAVC is H.264 but that is a slight over simplification I would hazard to suggest.
Yes Sony market it aggressively. They spent millions developing the advanced pre-encoding, encoding and decoding involved in the XAVC family. To get ratification by SMPTE the performance of XAVC was examined and tested to the extreme. Once its performance criteria was proven it got ratified. The only other H.264 based camera codec that has met this extreme SMPTE ratification process is Panasonic's AVC-Intra and Long GOP codecs.
To say XAVC is H.264 and the rest is marketing hype is akin to saying a family sedan and an F1 racecar both have engines. Yes this is factually correct. What is missing from this rather brief statement is any factual analysis. Yes both the family sedan and the F1 car have engines (H.264) but the design, execution, application, performance parameters and power delivery of the two cars is totally and utterly different. Even a brief study of how Panasonic's and Sony's H.264 implementations work at their different Profiles and Levels in AVC-Intra and XAVC-I would reveal to an even mildly interested party the these two H.264 developments are streets away from the base MPEG-4 Part 10, Advanced Video Coding (MPEG-4 AVC) H.264 implementation.
There is no reason why Panasonic or Sony would have gone to the expense of developing their families of H.264 and getting them passed and accepted as film and broadcast standards if they could have just used good old "H.264" MPEG-4 Part 10 MP4. Nor would now some 30 odd film and broadcast industry partners have bothered to sign up to licence agreements to use these two families of H.264.
With the Sony XAVC format it complies with the H.264 format specifications, agreed. Though Sony has strived to improve the image quality, while maintaining file interoperability with products provided by other manufacturers. The XAVC bit stream is based on a traditional KLV structure. One key element of Sony's development on this bit stream is the frame-by-frame integration of SPS (Sequence Parameter Set) and PPS (Picture Parameter Set) introduced into that bit stream. This allows the recording device to dynamically optimize the image quality per picture frame, following that the subsequently optimized image setting value will remain attached to the picture file after editing. It also helps to optimize the image quality during random access playback.
As previously mentioned the XAVC encoder has a pre-coding mechanism, which helps to make maximum use of that data in that it allows pre-analysis of each image frame. This pre-coding mechanism is integrated into both software and hardware encoders. This two-stage encoding process also takes place during high-frame-rate recordings and 4K recordings as well which is something totally outside the capabilities of base MP4 MPEG-4 Part 10, Advanced Video Coding (MPEG-4 AVC). Below is a list of ratified XAVC flavors. All miles away from MPEG-4 Part 10.
So yes you are right XAVC is H.264 but that is where the similarity ends.
Chris Young
Sony's latest APS-C model A6600 advertises itself as having a 16-bit still converted to the 14-bit. One has to assume that this makes the final image closer to the 16-bit (and that's from a $1,400 camera!) than to the 14-bit.
And, while the video is only 8-bit, it looks pretty good and I wonder if a similar 10-to-8 conversion is taking place in camera.
And, while the video is only 8-bit, it looks pretty good and I wonder if a similar 10-to-8 conversion is taking place in camera.
cyvideo
Veteran
Sony may quite well be doing something like you suggest. They always seem to be tinkering on the edge of their developments. Just looked at some of the stills out of the 6600 on DPReview. This one in particular renders a very nice looking sky and it's blue without some of the cyan overtones you often see in certain images. Punch it up and have a look. A very nice 8-bit JPG.
Chris Young
https://www.dpreview.com/sample-galleries/9344804222/sony-a6600-sample-gallery/0394521233
Interesting. And in that same gallery - you can see how "those awful Sony skin tones " (I'm paraphrasing others) and "horrible highlight retention" make for such a bad image. Look away - look away. :-KnockedOut(D
https://www.dpreview.com/sample-galleries/9344804222/sony-a6600-sample-gallery/4079738396
https://www.dpreview.com/sample-galleries/9344804222/sony-a6600-sample-gallery/4079738396
I understood you to be saying graphic cards and 8 bit panels had a role in the emasculation. My bad.
A few days ago I stumbled onto an A7RIV review (of sorts), with the talking head video shot on the same A7 RIV. Normally, it is not deemed a videocentric camera, what with the 61 MPX sensor and all (yes, it has an S35 crop). But, I thought that the colors were fine and the image was really good for the hybrid shooters.
combatentropy
Veteran
But, I think, in the olden days, a lot of data was just tossed out before it was compressed. Now, they don't have to do that unless it's a full sensor readout with a lot of shmegapixels.