C300 Discussion

[Brian Drysdale]

In this case and following a discussion I had with a Sony rep about the F3 not having a broadcast standard codec on board, I suspect one of the reasons is keeping the power requirements of the camera down. The RED cameras are power hungry and you really need to keep on top of the your batteries. The Epic appears to be less demanding than the RED One, but it is one of the considerations for the designers.

With the points made in Emmanuel's link, the C300 appears to be one of a family of cameras. The often referred to sticking point is the price, so how it and the family fit in with the competition remains to be seen. Until it's been tested and used on productions, we won't know if the C300 is worth the premium over the base F3.

RED keep their camera costs down because you're more or less buying directly from the factory warehouse, rather than a chain of dealers. There have been a number of products that have done this in the past.

Someone may know this area better, but I suspect the TV transmission chain is 8 bit, so a couple of the bits mightn't be used on 10 bit displays, although still giving high quality with what they're getting.

[Brian Drysdale]

Originally Posted by Emmanuel Plakiotis

Now that we are over the bit argument, a link with some interesting info for the camera:

Here's Canon's managing director being interviewed.

Interview with Canon Managing Director Mr. Maeda | Film and Digital Times: News

[David Heath]

Originally Posted by Don Miller

It's interesting that 4K has no more real information than the 1080p output. Which is a backwards explanation of why the C300 image could/should be superior to the 2-3K bayer image like the F3.

That's not what I was saying. (And I think you've attributed one of my quotes to Brian?)

At the root of all this is the fact that "4k" has become a very misused term. Marketing people like short snappy quotes - so "our camera is 4k!" is great to them. Accurate technical descriptions are what they hate - people switch off.

The most important point is to distinguish between a "4k system" resolution and 4k sensor resolution. (Sorry, all you marketing people!! :-) )The former implies full 3840x2160 luminance resolution (see note below), and chrominance dependent on the colour space. A 4k sensor will have 3840x2160 photosites - half green, quarter each red and blue - but that will not, CAN NOT, give full 4k system resolution. It will give it's maximum with a proper deBayer - something of the order of 80% (?) of 3840x2160 for luminance, less for chrominace.

The significance of what Canon are doing is saying that for a 1080p output, we don't need 80% of 4k - just 50% of 4k is 100% of 1080p. And we can get that standard SIMPLY via direct read out. "Simply" here means with little processing, no downconversion, scaling etc. That means lower power consumption, and no downconversion is likely to make alias charecteristics more like you'd expect from a 3 chip design.

It's not true to say that the 4k sensor has no more inherent information than 1080p. What is true is that Canon are choosing to say "with 1080 recording, we'll ignore that resolution *at the readout stage* for the sake of simplicity, that will still give very good 1080, and true 4:4:4 at that". In this case, "simplicity" needs to be seen as a good thing. It's also refreshing to see a manufacturer being so open, and I suspect it's because they know they're on to a good thing.

[NOTE - "4k" doesn't even have an exact meaning in terms of numbers - it's a family rather than a unique specification. I've used the QFHD (Quad Full High Definition) form of 4k for above (3840x2160) as that's most relevant to the C300 here. Note that the full sensor has a lot more photosites than currently being used, which could be used in a future camera for other forms of 4k.]

[Henry Coll]

Originally Posted by Brian Drysdale

With the points made in Emmanuel's link, the C300 appears to be one of a family of cameras. The often referred to sticking point is the price, so how it and the family fit in with the competition remains to be seen. Until it's been tested and used on productions, we won't know if the C300 is worth the premium over the base F3.

Yes, this is an interesting point.

Larry Thorpe has said that the C300 will have "brothers and sisters" in a few years. So given the current specs and the typical Japan tradition of incremental upgrades, just one at a time, I wonder how those might be.

The C300 might have a better or worse sensor, we don't know yet. But the C300 is actually an FS100 kind of camera. It's better built, has a Log curve and an additional SDI, but for the rest it's just like the FS100, only the latter also includes 60p and AF.

The question is how much will be a future model with 60p, 3Dlink, LUTs and a better body (actually F3 features) when the C300 is already $20k. Let's not forget Sony might introduce an F4 sooner, perphaps as soon as NAB2012.

And which features will be taken out to make a camera in the FS100 price range? Let's remember that price range is the one that sells in the hundreds of thousands and it's the most profitable segment. Will it be 7bit? (sorry, couldn't resist)


As for 4K , who really cares? Ask any producer and they'll tell you:

-It makes everything way more expensive
-Everything from Terminator2 to Avatar has been done at HD/2K
-There's no distribution channel for it, and won't be for more than a decade or two

[Paul Curtis]

Scooting through the latest posts i wanted to add something.

The C300 claims to produce an 8 bit log output - a flatter version suitable for grading. This isn't in camera but the output image is log.

In a typical 8 bit system the colour divisions (256 of them) are distributed linearly through the image from shadows to highlights. But that's not the way light works, light is exponential and a typical 256 bit image wastes values either in highlights or shadows. Now it depends what kind of log file the C300 produces but the potential is that it allocates those 8 bits to the areas within the image that require then. If it's an S log type curve then we should get better shadow and highlight roll off.

Yes the wrapper is 8 bit, easy to deal with, grade etc,. One graded then the final output would most likely be 10 bit+ because that will be linear.

Most bayer HD sensors don't have as much resolution as say a 3 chip 1080p camera because the sensor is dealing with colour recreation etc,. The 4K internally, as mentioned before, would produce a real 1080p resolution image.

If you look at something like the Scarlet. The filter is quite soft (if it's the same as the Epic) and you're probably going to be running in 4k or even 3k which is a windowed area of the sensor. So the resolution and field of view will suffer accordingly. If you want high speed, you'll be at 3k which is like 16mm. To cut that with 4k you'd need to move the camera or change lenses.

Without empirical evidence it's difficult to verify.

I would expect the C300 HD image to have more resolution than the scarlet and perhaps the F3/FS100

cheers
paul

[Brian Drysdale]

Originally Posted by Henry Coll

The C300 might have a better or worse sensor, we don't know yet. But the C300 is actually an FS100 kind of camera. It's better built, has a Log curve and an additional SDI, but for the rest it's just like the FS100, only the latter also includes 60p and AF.

With it's internal broadcast accepted codec it's better than the FS 100, However, the C300 seems to be targeting the F3 ^why hasn't it got an on board broadcast codec" market, but the current pricing is placing it against the s-log version of the F3 in the selection process.

To succeed it needs to demonstrate it has positive advantages over the other camera. Perhaps one may be the recording media, which reduces the running cost difference over time.

[Don Miller]

Originally Posted by David Heath

That's not what I was saying. (And I think you've attributed one of my quotes to Brian?)

I was commenting, not summarizing what you were saying. The C300 acquires the same amount of real information recording in 1080p or 4K, except for what is lost in compression.
At 1080p at each pixel the C300 has full RGB plus a complete luminance measurement. The F3 at each pixel has color and luminance for one color, and estimates luminance and the two remaining colors from the neighboring pixels.
Unless the F3 has far superior read noise performance compared the C300, the Canon puts considerably more information into a 1080p file.
It may not matter if the Canon is shot with an external recorder at 4K, or a 4K projector simply interpolates C300 1080p by adding lines. Either way has the same amount of real information. This is likely also the reason C300 1080p output looks good on the big screen. There isn't the compromise of a 2K bayer sensor.
Red doing 1080p from their high res bayer sensor is more interesting comparison to the C300. Mathematically there should also be less error in building individual pixels in Red compared to the F3. If Red runs the demosaicing algorithm on the PC it may be quite sophisticated. Both Red and the C300 have considerably more real data to work with than the F3.
We may be at the point where web compression is inadequate to see differences.

[Don Miller]

Originally Posted by Paul Curtis

In a typical 8 bit system the colour divisions (256 of them) are distributed linearly through the image from shadows to highlights. ..........

Is that unique to 8 bits? My understanding is different at the capture end. I though even without log the relationship between luminance and bits was geometric - brighter has more bits per stop.
In editing 8 bit Canon log it might be beneficial to dither to 10 bits.

[David Heath]

Originally Posted by Don Miller

I was commenting, not summarizing what you were saying.

In post no 313, you quote something I wrote (post no 308) - but attribute it to Brian Drysdale.

The C300 acquires the same amount of real information recording in 1080p or 4K, except for what is lost in compression.......and .....It may not matter if the Canon is shot with an external recorder at 4K, or a 4K projector simply interpolates C300 1080p by adding lines. Either way has the same amount of real information.

No, that's not true. The way the C300 does it is easy to understand. You effectively get the same as a 3-chip camera with 3 1920x1080 sensors. Effectively, full 4:4:4 1080p. Compared to the full chip, it's 50% resolution for luminance and chrominance.

But if you take the whole sensor and do a "true" deBayer, you'll end up with a 4k (3840x2160) raster - the question is what the "real" resolution will be. The actual figure will depend very heavily on the deBayering algorithm used, but something like 80% for luminance may be a decent ball park figure. Hence something like 3000x1725 - hopefully better with a good algorithm.

http://www.stark-labs.com/craig/arti...yering_API.pdf gives some good detail, and compares a poor (but very easily processed) algorithm ("nearest neighbour") with more complex ones giving better results. See why the 80% is such an approximation?

At 1080p at each pixel the C300 has full RGB plus a complete luminance measurement. The F3 at each pixel has color and luminance for one color, and estimates luminance and the two remaining colors from the neighboring pixels.

But the F3 doesn't have a 1920x1080 Bayer sensor. See my previous post (no 293, page 20), it's about 2456x1372 effective. Hence, using the 80% estimate, I'd expect output luminance resolution of around 1950x1100 - just what you'd want pre-downconversion! See why Sony chose the 2456x1372 figures? :-)

For 1080 output, there shouldn't be much difference in perceived resolution between the two - but the C300 will be able to make it with far simpler processing. Another sign of that should be if you compare the difference in power consumption between the two cameras.

[Roger Shealy]

This writeup from Adam Wilt provides some good details on the C300

Quick Look: Canon EOS C300 LSS 1080p Camcorder

[Jon Fairhurst]

Regarding the bit depth, it shouldn't be controversial.

* Yes, 10-bits is more than 8 and carries more information.
* 8-bits is enough for many productions.

The real question is what YOU want to achieve with it. If you are a heavy grader, every time you see contours, you'll wish you had more bits. If your style is more modest grades, you'll be perfectly happy with 8-bits.

It's clear that using an existing chip is the reason for 2K and 8-bits. Canon had a choice: release an 8-bit cam or delay the launch by months, a year, or more. They chose 8-bits, now. For this particular product.

The question isn't whether 8-bits is enough for the market. Is 8-bits enough for you, your shooting and processing style, and your end product?

[Chris Hurd]

Originally Posted by Jon Fairhurst

It's clear that using an existing chip is the reason for 2K and 8-bits.

Just to clarify Jon's post: existing chip = Digic DV III image processor.

[David Heath]

Originally Posted by Roger Shealy

This writeup give some good detail on the C300

Quick Look: Canon EOS C300 LSS 1080p Camcorder

If you only have time to read two articles about the C300, they should be the above and Chris's initial write up, linked at the beginning of this thread. As usual, it's hard to fault anything Adam Wilt says.

It may be worth making one small clarification though. Adam says:

The super35mm-sized, 24.6 x 13.8mm sensor is “quad HD”: 3840x2160. The CMOS sensor uses an RGB Bayer-pattern ......

The actual sensor is 4206 x 2340 (according to the Canon website) - though the actual read photosites are less, 3840x2160. In the case of the C300, the difference may be academic - BUT it's a very strong pointer that we are likely to see a more advanced model.

QFHD (3840x2160) is only one "4k" resolution (see 4K resolution - Wikipedia, the free encyclopedia ) With the sensor Canon have developed, it's capable of more than QFHD, capable of other forms of 4k. For example, from the wikipedia link: "In July 2010, YouTube began streaming certain videos at 4096 × 2304 pixels (in the 16:9 aspect ratio)....."

I'm also relieved to find that I and Adam are in broad agreement over true deBayer resolution. (Page 2, scroll about half way down.)

2x2 decoding in this manner gives you resolution measured in TVl/ph limited to 50% of the photosite-per-scanline count, whereas sophisticated deBayering gives you about 80% of the photosite-per-scanline value.

Well worth reading Adams explanation of what the camera is doing and how it compares with normal Bayer sensors.

[Jim Martin]

Originally Posted by Dom Stevenson

Chris Hurd at 3.41 pm

Thank heaven's for that. Some sanity at last. Great post!

I second that, Thanks Chris....

Jim Martin
Filmtools.com

[Brian Drysdale]

Originally Posted by Roger Shealy

This writeup from Adam Wilt provides some good details on the C300

Quick Look: Canon EOS C300 LSS 1080p Camcorder

Interesting article, although I would say that the shoulder rig shown is the most horrendous piece of industrial design I've ever seen. Fortunately, it's early days and I hope the 3rd party manufacturers will come up with a more elegant piece of modular design, rather than an improvised piece.

Canon aren't alone in this, it's surprising how many threads there are in forums about this every time a camera comes out. Yes it's modular, but the ergonomics of how the modules fit together should be part of the design. It's nothing new, the Panaflex is modular and it goes back to the early 1970s and there's a lot of hand held in "Jaws" on the boat. It's heavy, but balanced.

This is more of a problem when the camera is kitted out with 35mm cine lens accessories than in the neat looking configuration with the still lens when it's more Canon Scoopic