C200: What stops the C200 from being a near perfect camera?

So, help me with this (again) - let's assume that a given camera throws away a lot of info to get to the desired bit rate. That's where you get the 4:2:0 and so on. Now, let's look at Raw. If one has it in post, then tossing away info is unnecessary. An editing/file conversion program could then theoretically do a dual/triple/quadruple debayering/demosaic job, if need be. And then it becomes a question of time related to the processing power of the editing station. In that sense, it's sort of/kind of like an up-res, it's just that the debayer/demosaic process counts the pixels that are already there.

Right?
 
Is our poster suggesting the raw might be wonky? Fair point we don't know what the compression will do. I once shot some stills using 'S-Raw', whatever the hell that is, on a Canon.. and they were awful, nearly had to reshoot.

However at 1gbs it will probably be a lot less damaged than most of the codecs we know.
 
Yes, all discussion is of course based on the assumption that we're talking about 'normal' raw, albeit compressed here.

1Gbit/s is consistent with a 4096x2160 raw readout with around 5x compression.

Once a manufacturer starts inventing new names for their raw codec, there is always the risk that they came up with a great idea to skip a pixel here or there, or to invent something completely new like "Long GOP- but now in raw!"

But looking at some examples where people have posted working with the raw files, there is no reason to suspect foul play. I think it's simply the raw data compressed and then packaged to provide just the amount of flexibility that Canon sees fit to offer.
 
Wow, lots of misinformation flying here. Simply put: raw is not 4:4:4, nor is it 4:2:2, nor is it 4:2:0. Raw is raw sensor pixels. That is all it is.

Any single-sensor 4k Bayer-pattern sensor is going to deliver 2048x1080 red, 2048x1080 blue, and two 2048x1080 Green matrices. That's it. That is what raw is. Doesn't matter if it's from Sony or Canon or Panasonic or Red or BlackMagic or any other manufacturer.

If you want 4k 4:4:4, you would require 4096x2160 red, 4096 x 2160 blue, and 4096 x 2160 Green pixels. That is the definition of 4K 4:4:4. No single Bayer-pattern sensor can deliver that. It is impossible. It would require three 4K chips, it can't be done with one. If you wanted to get true 4k sampling on a single sensor, it would require an 8k sensor that would then downsample.

That's just how it is. There is only one red sample for every four sensor pixels, and there is only one blue sample for every four sensor pixels. If you are recording that data as a so-called "4:4:4" file, it will mean uprezzing 2048x1080 red pixels to fill in a 4096x2160 frame. There is no way around it. That's all the color samples the Bayer sensor can deliver.

4:2:0 is decimation that happens in the recording, not at the sensor. But when a 4k signal is decimated to 4:2:0, guess what you get: 4096 x 2160 in brightness, and 2048 x 1080 in color resolution. Which is all that a Bayer sensor can SAMPLE, which is why I said you won't get 4K 4:4:4 in raw. You'll get substantially less. After the demosaic process it will be somewhere around 3.2k of resolved detail, with about 2048x1080 worth of color resolution. Anything more would not be sampled resolution, it would be sophisticated synthesization.

Doesn't make it "bad", it's just what it is, regardless of who is doing it. Ain't no way to get a fully -resolved 4k 4:4:4 from a Bayer pattern single sensor. You can up-res the color channels (which is a gross oversimplification of what the demosaic process does), but that isn't sampled resolved detail.

Again, if you want actual genuine 4k 4:4:4, you'd need three chips. A single sensor would need massive oversampling to provide a true resolved 4k 4:4:4 (at least an 8k sensor). A single 4k sensor can make some gorgeous images, sure, but they will not have the resolved detail of actual 4k 4:4:4. That's all I was trying to clarify.
 
Which is why I wish Canon would release their 8K sensor so that we could get an actual near 4K resolution that would allow a close to 1:2 punch in for interviews for 1080p.
 
Bassman2003 said:
The C200 has an 8.85MP sensor.
Click to expand...

Which means it has the biggest pixels possible for an S35 4K sensor. It should excel in dynamic range, sensitivity, and noise performance. But it won't be the sharpest 4K camera in terms of resolved detail, nor will it come close to sampling 4K 4:4:4. You'd need an 8K sensor (at least) to deliver fully -resolved maximum sharp 4K at 4:4:4. And that would likely mean an 8,192 x 4,320 pixel matrix, which would mean about a 35-megapixel sensor.

(Which would mean less dynamic range and perhaps less sensitivity, although downsample should negate the increased noise to be expected when using smaller pixels).

There's no free lunch and you can't have it all, especially in the lower price brackets. Canon appears to have prioritized sensitivity, low noise, and dynamic range. Ain't nothing wrong with that and should make customers very happy. All I was responding to was a claim that seemed to be saying that people are expecting 4K @ 4:4:4 from this camera, and that's just not possible using a single 4K Bayer sensor.
 
Barry_Green said:
Wow, lots of misinformation flying here. Simply put: raw is not 4:4:4, nor is it 4:2:2, nor is it 4:2:0. Raw is raw sensor pixels. That is all it is.

Any single-sensor 4k Bayer-pattern sensor is going to deliver 2048x1080 red, 2048x1080 blue, and two 2048x1080 Green matrices. That's it. That is what raw is. Doesn't matter if it's from Sony or Canon or Panasonic or Red or BlackMagic or any other manufacturer.

If you want 4k 4:4:4, you would require 4096x2160 red, 4096 x 2160 blue, and 4096 x 2160 Green pixels. That is the definition of 4K 4:4:4. No single Bayer-pattern sensor can deliver that. It is impossible. It would require three 4K chips, it can't be done with one. If you wanted to get true 4k sampling on a single sensor, it would require an 8k sensor that would then downsample.

That's just how it is. There is only one red sample for every four sensor pixels, and there is only one blue sample for every four sensor pixels. If you are recording that data as a so-called "4:4:4" file, it will mean uprezzing 2048x1080 red pixels to fill in a 4096x2160 frame. There is no way around it. That's all the color samples the Bayer sensor can deliver.

4:2:0 is decimation that happens in the recording, not at the sensor. But when a 4k signal is decimated to 4:2:0, guess what you get: 4096 x 2160 in brightness, and 2048 x 1080 in color resolution. Which is all that a Bayer sensor can SAMPLE, which is why I said you won't get 4K 4:4:4 in raw. You'll get substantially less. After the demosaic process it will be somewhere around 3.2k of resolved detail, with about 2048x1080 worth of color resolution. Anything more would not be sampled resolution, it would be sophisticated synthesization.

Doesn't make it "bad", it's just what it is, regardless of who is doing it. Ain't no way to get a fully -resolved 4k 4:4:4 from a Bayer pattern single sensor. You can up-res the color channels (which is a gross oversimplification of what the demosaic process does), but that isn't sampled resolved detail.

Again, if you want actual genuine 4k 4:4:4, you'd need three chips. A single sensor would need massive oversampling to provide a true resolved 4k 4:4:4 (at least an 8k sensor). A single 4k sensor can make some gorgeous images, sure, but they will not have the resolved detail of actual 4k 4:4:4. That's all I was trying to clarify.
Click to expand...
What is the F65 in relation to this?
 
Barry_Green said:
You'd need an 8K sensor (at least) to deliver fully -resolved maximum sharp 4K at 4:4:4.
Click to expand...

But once you have an 8K bayer sensor, you will look at it and say, you know this sensor doesn't give me 8K 4:4:4. What I need is 16K. And so on. It never ends. Barry, don't feed them!

The thing we have to remember is that 4K is an arbitrary number. There is nothing magical about 4K.

There was something magical about 2K. 2K is about what you need to outresolve the human eye, at normal screen distances (including movie screens). When they developed HDTV, they were taking that into consideration. I remember reading about it in Popular Science in the 1990s.

But 4K is just taking the logically derived 2K and doubling it. There's no reason you have to multiply 2K by 2.0. You could have multiplied by 1.5, or by the square root of 2, and still achieved an upgrade (and why did you need that upgrade, considering that it was already at the level of the human eye?). But no. You had to just go and double it.

Okay, 2K, you say, was skating too close to the edge. Sure, it matches the human eye. But what if circumstances aren't ideal? What if it falls to 1500x800, because I had to crop it for some reason, or because of video compression? We all know that real 2K looks great, but 1500x800 looks like your uncle's old VHS camcorder. (Where is my sarcasm font?)

Okay, well, with even 4K 4:2:0, I think you have what is needed for the human eye and plenty of margin.
 
Fishflavour said:
What do you think is stopping this from being a near perfect camera?
Click to expand...

Probably its imperfections.




Sorry, the discussion got way too techy - so I thought I'd give the simplest answer possible to the original question.

With all this talk about 444 vs 422 vs 420 vs RAW and de-Bayering I think it's important to do an apples to apples comparison. The fact is all the cameras in this price range (and a lot higher than this price range) use the same tricks and the same terminology to reach the same quoted specs, however arbitrary those specs or marketing terms are.
 
Last edited:
Also do we really know what goes on to achieve these impressive figures? It doesn't mean much, the Alexa Mini doesn't shoot true 4K and apparently it's not on the Netflix approved camera list for content but I know I would still choose an Alexa over and above all the other cameras out there if money was no object.

The 2K Arri image regardless of 4K DCI pisses on mostly everything else when it comes to digital cinema full stop.
 
combatentropy said:
The F65 uses a quincunx pattern, which I thought was kind of cool (Q67: Bayer with a twist).
Click to expand...

With Q67 standing for "Quality 2/3". i.e., or 17.6 MPX in 4K 4:2:2. A 4:4:4 true RGB 4K would be 8.8 x 3 or 26.4 MPX (in a traditional Bayer pattern). Red Weapon Helium sensor is 35.4 MPX, which they present as 8K.

PS. The eye can't tell 4:4:4 from 4:2:2, which is one of the reasons that the Bayer pattern was created.
 
Barry_Green said:
Wow, lots of misinformation flying here. Simply put: raw is not 4:4:4, nor is it 4:2:2, nor is it 4:2:0. Raw is raw sensor pixels. That is all it is.
Click to expand...

Forum life can be hard at times. First of all, let me just say that I agree with everything you say about Bayer sensors and traditional encoding into the various chroma subsampling flavours.

One last stab at this. I'm maintaining a single chip 4k CMOS sensor with a Bayer pattern as our reference. As we all agree on and are aware of, that sensor can be broken down into 4 matrices of pixels: one Red, two Green (offset from each other) and one Blue. These matrices are each '2k' in resolution due to the "holes" of missing pixels in them, according to the Bayer pattern. Just like you say above.

Canon's image processing pipeline in the camera (for 2k/HD—not 4k this time) includes extracting these matrices and extrapolating the data by filling in the holes (mostly in an effort to avoid aliasing) and then recombining them into a RGB 444 component. But this makes sense, right? This is an example of a 4k Bayer sensor delivering 2k 444. This is what you are saying is correct, and I agree.

But then we have the raw scenario. I kind of agree that "raw is raw", but in practical terms "raw" differs quite a bit between camera brands. We don't (ever?) get the RAW image data, but it's packaged somehow. Canon's solution to this is to take all the matrices mentioned above (that carry all the color information the camera can deliver) and send them out as four parallel data output streams—as opposed to a single raw data stream.

Each matrice is then structured as a separate file that include audio, metadata and management files. These, in turn, are then multiplexed to form the final raw output of the camera as a single stream. Canon describes this output as Bayer RGB 4444.*

By doing this, and now I'm more speculating, Canon allows the user to offset/stretch/compress these individual files to change overall contrast and white balance.


* = Had I used this more proprietary term that Canon uses from the beginning, I might have saved this thread some confusion and myself some embarrassment. It's what I alluded to in a previous post when I talked about using the camera maker's own labelling of their output, but I didn't have the correct wording on the tip of my fingers. I simply wanted to say that the raw output from the C200 should extract ALL the color information the sensor can offer, at 12 bit. I'm happy with that.
 
You must log in or register to reply here.
Back
Top