Something new from Canon on Nov. 3rd...

[Lawrence Bansbach]

Originally Posted by Jon Fairhurst

* Give me more than eight bits of depth (along with a tall dynamic range)
* Reduce coding artifacts so I can translate that bit depth into smooth, graded images.

I find it odd that Canon is sticking with MPEG2, when AVC seems so much more efficient. So far, only Panasonic has announced any plans to implement a codec that should meet your criteria. AVC Long-GOP will support 24p, 25p, and 30p at 10-bit 4:2:2 at data rates up to 50 Mbps. It's too bad Panasonic didn't extend this to 60p at 80-100 Mbps. If they don't include AVC Long-GOP in the GH3 and maybe a cheaper, VG-20-level video camera, well, I'd be sorely disappointed.

[Jim Martin]

I'm not sure why they would change.....their codec is a broadcast codec, approved by the BBC & Discovery HD, and basically equal to Pro-Res422....the same codec that is in the Arri Alexa.

Jim Martin
Filmtools.com

[David Heath]

Originally Posted by Lawrence Bansbach

I find it odd that Canon is sticking with MPEG2, when AVC seems so much more efficient. So far, only Panasonic has announced any plans to implement a codec that should meet your criteria. AVC Long-GOP will support 24p, 25p, and 30p at 10-bit 4:2:2 at data rates up to 50 Mbps.

Well, the AVC nature means a bit more efficiency, but it's difficult to implement it well in real time, so the datarate saving may not be worth it. By AVC Long-GOP, do you mean AVC-HD? If so, that doesn't come in a 10 bit or 4:2:2 version, nor at 50Mbs. The more upmarket Panasonic codec is AVC-Intra, and the 100Mbs variant is seen as approximately equivalent to XDCAM422 - but that's not long-GOP.

What I personally would like to see is the ability to record EITHER XDCAM422, OR a RAW mode directly from the sensor. The former would then be preferable when time is important (and it's still a broadacst quality codec), the RAW mode would be far preferable for long term projects with the expectation of a long time spent grading etc, and be far the best for keeping all options open. The same principle as being able to save JPEG and/or RAW with a DSLR.

The advantages are less down to compression quality - far more being able to have a lot more post control.

[Jon Fairhurst]

Canon's DSLRs use h.264 (otherwise known as MPEG-4, Part 10, or AVC) rather than MPEG-2.

And, FWIW, RED compresses the RAW signal with Wavelet coding. I could live with that. ;)

[Lawrence Bansbach]

Originally Posted by Jim Martin

I'm not sure why they would change.....their codec is a broadcast codec, approved by the BBC & Discovery HD, and basically equal to Pro-Res422....the same codec that is in the Arri Alexa.

I was under the impression that Canon's MXF format is 4:2:2 but only 8-bit, whereas Pro-Res422 supports 10-bit. But beyond that, H.264, being more efficient, would probably look better at 35 Mbps than MPEG2 at 50 Mbps (of course, that's an oversimplification). And lastly, Canon's HDSLRs are prosumer, and most consumer and prosumer cameras are trending to H.264.

Originally Posted by David Heath

Well, the AVC nature means a bit more efficiency, but it's difficult to implement it well in real time, so the datarate saving may not be worth it. By AVC Long-GOP, do you mean AVC-HD? If so, that doesn't come in a 10 bit or 4:2:2 version, nor at 50Mbs. The more upmarket Panasonic codec is AVC-Intra, and the 100Mbs variant is seen as approximately equivalent to XDCAM422 - but that's not long-GOP.

Panasonic recently announced details of its AVC-Ultra "family" of codecs. One is AVC Long-GOP which is indeed 10-bit 4:2:2 at up to 50 Mbps. Of course, nothing has been released sporting the new codec, but I expect the AF100's successor will do so, if not AVC-Intra at up to 200 Mbps. They'd be stupid not to include AVC Long-GOP in the GH3 and possibly also a VG-20-style camera. As to XDCAM422, my understanding is that it is, like Canon's format, only 8-bit.

[Peter Moretti]

Originally Posted by Jim Martin

I'm not sure why they would change.....their codec is a broadcast codec, approved by the BBC & Discovery HD, and basically equal to Pro-Res422....the same codec that is in the Arri Alexa.

Jim Martin
Filmtools.com

Yeah, but Jim, the codec in the Alexa that everyone drools over is the 4:4:4:4 codec.

And isn't 4K four times the data of 1080? Is 50Mbps really enough?

[Glen Vandermolen]

Originally Posted by Peter Moretti

Yeah, but Jim, the codec in the Alexa that everyone drools over is the 4:4:4:4 codec.

And isn't 4K four times the data of 1080? Is 50Mbps really enough?

50Mbps is good enough for BBC HD and Discovery HD. That says a lot.

Keep in mind the Alexa is approaching the $100K mark. Do you want this new Canon (if there really is a new Canon) to compete in that price range?
A large sensor, interchangeable lens, 50mbps, 4:2:2 codec video camera for under $10 grand sounds really good to me. 10 bit would be nice, too.

[David Heath]

Originally Posted by Lawrence Bansbach

Panasonic recently announced details of its AVC-Ultra "family" of codecs. One is AVC Long-GOP which is indeed 10-bit 4:2:2 at up to 50 Mbps. Of course, nothing has been released sporting the new codec, but I expect the AF100's successor will do so, if not AVC-Intra at up to 200 Mbps.

Ah, OK, found this - AVC-Intra - Wikipedia, the free encyclopedia . AVC-Ultra has been talked about for a while, but I'd only heard it referred to an I-frame only version at 200Mbs - presumably more intended to rival HDCAM-SR. They now seem to have widened the meaning, so the version I was aware of is now "AVC-Intra Class 200".

According to wikipedia, what you're referring to is:

AVC-LongG enables compression of video resolutions up to 1920x1080 @ 23.97, 25 and 29.97p, with 10 bits of pixel depth at 4:2:2 color sampling, at data rates as low as 25 Mbit / sec.

Well, we'll have to see. (But no standard for 1080p/50?) It's important to bear in mind that there's no magic here. All the current compression systems are fundamentally based on MPEG2 I frame only. That's extendable (to give better efficiency) by AVC techniques, long-GOP, etc to give the same quality at lower bitrates. The snag is that these extensions all require more computer power to process, so it's in to a law of diminishing returns. Why use the most complicated system if you can get the same result by just using a slightly higher datarate? Would you buy a car with highly advanced, highly efficient engine costing $10,000 more to get a fuel saving of $500 a year? There's an example when "efficient" doesn't always mean "sensible".

There's also the case of "good enough". It would be foolish design to engineer a camera with a codec far more capable than the front end is producing. "10 bit" for encoded video sounds "wow" on paper, and is great for marketing people, but it's only really worth it if the front end is up to it - otherwise you're just throwing 20% of your bitrate away on coding noise! And as regards your reference to the AF100, I suspect the Panasonic engineers realised that only too well, and that explains why it was only released with AVC-HD and not AVC-Intra. There will be little point in using a 10 bit codec with any successor unless that has a much lower noise floor - which will effectively mean a new designed-for-video sensor (as with the F3), not adapting one that was designed for stills.

Times will move on, but as Glen says, XDCAM422 is not bad for now.......

All this applies to cameras that record processed signals - video ready to be displayed and viewed. It's a completely different story when you talk about RAW - the unprocessed data effectively straight from the sensor photosites. In these cases 8 bit is nowhere near enough, probably not even 10 either. But the processes of de-Bayering, matrixing, gamma correction, gain, colour balance etc that then have to be done are so likely to raise the noise that it may be pretty pointless to output more than 8 bits after all this has been done.

The real reason for 10 bit is not that it will enable any big difference to be seen on the first generation, but that it provides more headroom for post processing - if the original signal is up to it. I'd rather skip that and go straight to some form of RAW recording for a camera of this type.

[Peter Moretti]

Originally Posted by Glen Vandermolen

50Mbps is good enough for BBC HD and Discovery HD. That says a lot.

Keep in mind the Alexa is approaching the $100K mark. Do you want this new Canon (if there really is a new Canon) to compete in that price range?
A large sensor, interchangeable lens, 50mbps, 4:2:2 codec video camera for under $10 grand sounds really good to me. 10 bit would be nice, too.

If it's a 4K sensor creating an HD frame size, then fine. But if it's creating a 4K frame size, then I don't see how 50 Mbps is enough.

And ten bits wouldn't be nice, it would be essential. Eight bits simply doesn't allow enough levels of gradation to record the full dynamic range of modern sensors out today.

[Peter Moretti]

Originally Posted by David Heath

...
All this applies to cameras that record processed signals - video ready to be displayed and viewed. It's a completely different story when you talk about RAW - the unprocessed data effectively straight from the sensor photosites. In these cases 8 bit is nowhere near enough, probably not even 10 either. But the processes of de-Bayering, matrixing, gamma correction, gain, colour balance etc that then have to be done are so likely to raise the noise that it may be pretty pointless to output more than 8 bits after all this has been done.

The real reason for 10 bit is not that it will enable any big difference to be seen on the first generation, but that it provides more headroom for post processing - if the original signal is up to it. I'd rather skip that and go straight to some form of RAW recording for a camera of this type.

David, you're ignoring the most important reason to use ten bits. Eight bits simply doesn't allow enough fine gradations to record more than about nine stops. And most modern sensors can record more than nine stops. So what happens is the cameras clip the highlights about a stop or two before the sensor itself clips, i.e. the pixels completely fill-up.

This is also why ten bits and S-Log provide real DR benefit to the F3.

[Matt Davis]

Originally Posted by David Heath

Well, we'll have to see. (But no standard for 1080p/50?)

At the other end of the AVC line, a new 2.0 spec was introduced in July 11, introducing AVCHD Progressive, along with AVCHD 3D and the inevitable AVCHD 3D/Progressive alphabetti spaghetti.

Progressive denotes the 50p/60p version - the AVC-Ultra codec may also have it's own variant thereof. I guess the wiki-weasel words 'as low as 25 Mbits' seems to hint that would be the absolutely lowest mode and you'd have others to play with (like the AVCHD line). Sigh, the nice thing about standards is that there's so many to choose from.

[Brian Drysdale]

All this depends on the market they're aiming the camera at. Recording the 10 bits on the F3 comes as an extra over the base price of the camera. If they have a 10 bit HD SDI on the new camera, how many people are going to go the extra mile if the camera has the broadcast accepted Canon XF 8 bit 4;2;2 codec on board?

Canon might do two versions: one version to compete with the F3 with its s-log and the Epic-S recording RAW REDcode and the other version to compete with the FS100 and AF100. These are two different markets, with differing budgets.

[Les Wilson]

Originally Posted by Monty Wentzel

... I need to really begin to perfect the art of story telling. Is there a forum for that?

Monty

I think you'll find some useful threads over in the Documentary Techniques section of DVinfo or start a new one there:
Documentary Techniques Forum at DVinfo.net

[Lawrence Bansbach]

Originally Posted by David Heath

The snag is that these extensions all require more computer power to process, so it's in to a law of diminishing returns. Why use the most complicated system if you can get the same result by just using a slightly higher datarate?

If it's so much more complicated, why is every manufacturer using H.264? I seriously doubt Panasonic would announce new codecs if it couldn't reasonably implement them. And I'm unaware of a commonly used 10-bit MPEG2 codec -- neither XDCAM422 nor Canon's MXF is 10-bit.

Would you buy a car with highly advanced, highly efficient engine costing $10,000 more to get a fuel saving of $500 a year? There's an example when "efficient" doesn't always mean "sensible".

Some people do it for the environment despite the added cost. As far as using a more complex codec, the analogy doesn't really hold. The car is mechanical, and I assume that, say, a complex hybrid engine and battery array are far more costly to manufacture than a conventional engine, at least for now. Camera codecs and associated processors are the products of constant technological evolution. By the time Panasonic implements AVC-LongG, it'll probably be as cheap to do so as AVC-HD is today.

There's also the case of "good enough". It would be foolish design to engineer a camera with a codec far more capable than the front end is producing. "10 bit" for encoded video sounds "wow" on paper, and is great for marketing people, but it's only really worth it if the front end is up to it - otherwise you're just throwing 20% of your bitrate away on coding noise! And as regards your reference to the AF100, I suspect the Panasonic engineers realised that only too well, and that explains why it was only released with AVC-HD and not AVC-Intra.

Possibly, but Panasonic claimed adding AVC-Intra to the AF100 would have added several thousand dollars to the cost.

[David Heath]

Originally Posted by Peter Moretti

David, you're ignoring the most important reason to use ten bits. Eight bits simply doesn't allow enough fine gradations to record more than about nine stops. And most modern sensors can record more than nine stops.

No, not true. What you are not taking into account are the gamma, knee etc processing which has the effect of compressing the dynamic range that the camera is capable of, into the dynamic range that the eye can see.

*AT ANY ONE TIME* the eye has a range of about 7 stops, but as you look from a shade area to something brightly lit the eye adapts. And the brain is clever enough to accept all this without realising what's going on, the impression is that the eye has a far better dynamic range. This is why 8 bits is enough for normal recording and viewing - it matches the range of the eye. It's easily proven by looking at greyscales and seeing how small the differences get before it merges into a continuous mass. (Corresponds to about 7 bits, or about 128 levels.)

So what happens is the cameras clip the highlights about a stop or two before the sensor itself clips, i.e. the pixels completely fill-up.

No. It may be that a doubling of light level in the mid tones may lead to a doubling of the recorded value. But as it gets brighter, the whole point of the knee is to introduce non-linearity, such that in the highlights a doubling of light level means maybe an increase in recorded value of (say) 25% - not 100%. Hence the headlights start to crush out gracefully (as with film) - not clip abruptly.

This is also why ten bits and S-Log provide real DR benefit to the F3.

S-log and 10 bits are indeed of benefit to the F3, but it's the combination that makes the difference - not either in isolation. The point is that the (say) 12 stops that the sensor resolves are now compressed not into about 8, but into 10. Looked at directly, it will look low contrast, flat and dull - but it's intended to be processed to give the effect of raising contrast - but allowing that to be done in a controlled way, not just as a camera set in a single way.

And it's important to realise the difference between a signal processed for direct viewing and recorded to 10 bit, and one processed to S-log and recorded to 10 bit - 10 bit has far more benefit in the latter case.

But even better than S-log and 10 bit is RAW, which really needs at least 12 bit to do it justice. In this case there's no dynamic range compression or knee at all, and no other processing at all. But it's no good for direct viewing. The comparison with film is that a negative may have plenty of detail in highlight and lowlight, but will need to be printed onto a more contrasty stock. Varying print exposure will give preference to either lowlights or highlights.

But the problem with S-log and RAW is that although they give plenty of control, they HAVE to be graded, which takes time. Not a problem for some work, but bad for others.

Hence that's why I'd like to see any new camera with the option of both - use as appropriate. Ideally, 1080p via XDCAM422, and 4k with a RAW system. Forget about 10 bit, it falls between the stools of quality and convienience. The real ideal may be two cameras - one with just the 1080p XDCAM422 option, the more expensive one with both.

We'll see.