Bolex D16: RAW, 16/Super 16 sensor size, costs less than MkIII

Dubdom said:
Hi Barry have you downloaded and looked at any of the ikonoskop dii sample images? They are DNG files from a bayer pattern ccd imager, in fact, this is the digital bolex in reality really - apart from the lack of 2k and the fact that they are more expensive. I think they even use similar imagers to the one feted to be in the digital bolex. I don't' think they are making any claims to delivering 4:4:4 though
Click to expand...
I haven't, no. I'm content enough that the DBolex will deliver a fine image.

My problem in this thread came up when HHL was all excited because the DBolex was claiming it was going to be delivering "2k 12-bit 4:4:4", and ... that's just not possible from a single Bayer sensor. That's the same problem I had with Red when they advertised the same thing; it just can't be done.

I just looked at the specs of the Ikonoskop dII, and it doesn't make any mention of 4:4:4, even though it is probably using the same sensor or one with comparable technology (1920x1080 pixels in a Bayer pattern). In that vein, I applaud Ikonoskop. And, in fact, the dII looks to be pretty much exactly the same thing as the DBolex, except that the dII is a bit more advanced -- 640x480 color viewfinder and an HD-SDI playback module built in, which would go a long ways towards explaining the price differential (well, that, plus the dII is actually a finished product and delivering now, whereas the DBolex is still in the prototype stage).
 
Speaking about the chroma subsampling of the dBolex, I just remembered this from EOSHD:

EOSHD: You have raw 4-4-4 but as far as I know, a bayer CCD sensor is by nature 4-2-2. Is there a reason for 4-4-4 on the specs list?

Elle: Our understanding from Kodak is that the chip is 4:4:4 in YUV and it may be 4:2:2 in RGB. We will look into this a bit more.
Click to expand...
 
vladszp said:
there are rumors that gh3 sensor will be global shulter and if this is true that will be ideal for full readout ...
Click to expand...
My problem with the "rumor" sites is that they almost universally have no idea what they're actually talking about. I would be flabbergasted if the $800 GH3 became the first available global shutter camera, when a $16,000 C300 with one of the most sophisticated digital imaging sensors on the market, and the $80k Alexa, both use rolling shutter.

af200(maybe onother name or class) will use tweaked sensor version similar as af100 ...
Click to expand...
That remains to be seen as well. Lots of people were convinced that the AF100 used the GH2 sensor, but Panasonic said it wasn't, and it isn't. Probably the same silicon wafer, but very different processing attached to each chip. I would certainly expect that if there are improvements made towards a GH3, then any future AF successor would have at least comparable improvements, sure.

RAW is simply ALL data from sensor nothing is left out ...
Click to expand...
Yep. Delivers 100% of what the sensor saw. Nothing less, and nothing more. And a Bayer sensor is incapable of resolving a true 4:4:4's worth of color information, so raw Bayer can't deliver more than what the sensor actually has.

is it 444 or 422 depends of sensor design 3 CCD or one (binning or not)...
Click to expand...
Well, the technical argument is that 4:4:4 applies to an RGB image, and 4:2:2/4:2:0/4:1:1 refer to the way an image processing system discards color information. But in the larger context, where a facial tissue is commonly called a "Kleenex" and a photocopy is frequently called a "Xerox", there's a usage of language that's imprecise but "we all know what we mean" when we say it. And in that context, whether technically precise or not, we mean in 4:4:4 that "there's an individual red, blue, and green color sample for every pixel" -- which is obviously a physical impossibility for a Bayer sensor. And when someone like Panavision refers to a Bayer sensor as "a 4:2:0 sensor", that is obviously incorrect terminology (and what Graeme objected to, most correctly). But in the larger sense of "well, what we MEAN by 4:2:0 is a situation where there's half as many color samples on the horizontal and half as many on the vertical", well, that's a reasonable definition of what a Bayer sensor delivers.

As of right now, the only way to get a true 4:4:4 is to either a) use three chips, or b) use a grossly oversampling sensor that has more than enough color samples in each direction for each color, and which then scales down to the target resolution.

The best example of that right now, IMO, is the C300's sensor. It's a Bayer pattern per pixel, but it's not a Bayer pattern sensor at all; it has 2x horizontal and 2x vertical oversampling so it delivers one red, one blue, and two green samples for each and every pixel. No de-Bayer demosaic is necessary, it's a true and legitimate 1920x1080x4:4:4 sensor. Only problem with the C300's sensor is that it's mated to a relatively-ancient three-year-old Digic DV III processor, which (while being Canon's latest motion image processor) is woefully out of date for handling what the sensor can give it. When/if they introduce a Digic DV IV with 10-bit 4:4:4 support, you'll really see what a camera with that sensor could do.
 
kylevant said:
Speaking about the chroma subsampling of the dBolex, I just remembered this from EOSHD:
Click to expand...
Well, glad he asked, but no, by nature, a Bayer sensor isn't 4:2:2. It isn't 4:X:X anything; that's digital processing notation, but if you wanted to call a facial tissue a "Kleenex" then you could say that a Bayer sensor is 4:2:0. Even though it isn't, it gets the basic idea across that you have half as many color samples in the horz and half as many in the vertical, even though applying that notation to a demosaic'd image isn't accurate -- any more than it's accurate to say that a 4K bayer sensor delivers a 4K image, because it most definitely doesn't, it delivers (at very best) a 3.2k image, so you can see where the confusion comes in...
 
reason to go global shutter will be to cut cost of camera production (no mechanical shutter) so they spend little more money on R&D but cut cost in production.... there are already small sensor companies who offer global shutter cmos ...
even Cypress company who produce arri alexa sensor design global shutter cmos... problem is loss of dynamic range(more transistor around pixel and less sensitivity )....
 
I'm trying to grasp the concept of true 4:4:4 from a sensor. In regards to the new JVC HMQ10, it records 4K (well, very close) through 4 separate streams into 4 separate SDHC cards, all from one sensor. Does this make it a Bayer sensor? Can you extract a 1920x1080 4:4:4 signal from the 4K video?
 
vanvideo said:
I'm trying to grasp the concept of true 4:4:4 from a sensor. In regards to the new JVC HMQ10, it records 4K (well, very close) through 4 separate streams into 4 separate SDHC cards, all from one sensor. Does this make it a Bayer sensor? Can you extract a 1920x1080 4:4:4 signal from the 4K video?
Click to expand...
If it's a single sensor, it has to be a Bayer pattern or some other color filter array arrangement (such as RGB Stripe, or the sort-of-Bayer that Canon uses). If it's a single 4K sensor (which it seems to be, the specs call it 8.2 megapixel, so roughly 4K x 2K) and it claims to output 4K footage, then it would likely be a Bayer pattern and would, in all likelihood, function somewhat similarly to the way the Red One does, meaning it'll output somewhere around 3K of actual resolution with a little better than 4:2:0's worth of color information.

Could you downscale it to 1920x1080x4:4:4? Probably, if you could get at the uncompressed data. But the biggest problem with the JVC that I've seen on the footage so far is that the codec they're using erases pretty much all fine detail, it looks as bad as the original GH1 "mud" problem, so ... would there be that fine detail available to downscale into 1920x1080x4:4:4? Doubtful. Not the sensor's fault, the compression's fault.
 
vanvideo said:
I'm trying to grasp the concept of true 4:4:4 from a sensor.
Click to expand...
Well, let's go back to this for a second. First thing to understand is: sensors are monochrome! You can't get color out of an image sensor, you can only get a monochrome image. So in order to get color, you have to do one of two things:
1) use three separate sensors, each with a primary color filter over them or by using a prism to optically split the light into its three primary colors, and have each sensor record a monochrome image composed entirely of that color, which you will then combine to create an R/G/B image, or
2) put individual color filters over individual pixels on the sensor, and then using a pixel's own color plus the color of the surrounding pixels you attempt to synthesize an approximation of what you think the original color detail was for each pixel.

Method 1 is the most accurate, highest-quality way to do it, and it's why all broadcast cameras use three chips, and is the pixel-accurate way to actually sample and store true 4:4:4 data. You have a legitimate and true red, green, and blue sample for every pixel in the image. Another way to look at this would be a Foveon sensor, where you have all three sensors "sandwiched" together.

Method 2 is the way it's done when you have only one sensor available (which is the case with all digital film emulation cameras). The color filter can be a Bayer pattern, or an RGB Stripe, or it can be the crafty new way Canon is doing it (which involves a bayer pattern at the individual pixel level, but not a bayer demosaic across the whole sensor). With a bayer pattern on the image, such as your proposed 3840x2160 JVC sensor, you'd have 1920x2160 green pixels, 1920x1080 red pixels, and 1920x1080 blue pixels. As you can see, in the context of a 3840x2160 image, you're not going to get 4:4:4 color out of only 1920x1080 red and 1920x1080 blue! But, there's enough red and blue that if you scaled the image down to 1920x1080, you could end up with a 4:4:4 1920x1080 image (again, though, this is dependent on having access to the raw data, not a compressed image where all the fine detail has been smushed out of it).

Canon's way of doing it in the C300 is to have a 4K sensor but only ever have aspirations of making a 1920x1080 image. So each final frame pixel is made up of a 2x2 block of two green, one red, and one blue. Every pixel of the final 1920x1080 frame has its own red, its own green, and its own blue pixel. It's a true 1920x1080x4:4:4 sensor, BUT -- that's all it can make, is 1920x1080. It can't ever make an image with higher resolution. (not a fault of the sensor, but it's the way they choose to read it out). Red, on the other hand, uses the exact same sensor configuration but does a de-bayer on the whole chip. That lets them get 3.2k of detail out of it, rather than the 2K Canon can get. But Canon gets 4:4:4, whereas Red gets, at best, a tad better than 4:2:0. Same number of pixels, same number of red/green/blue samples, but it's all in the way you process it and how you prioritize what you're trying to get out of it. It could be argued that Red's approach is more flexible, because if you took the true raw image and downscaled it to 1920x1080, you should be able to come up with an image that looks just as good as Canon's 1080px4:4:4, but you could never take Canon's 1080p and turn it into a legitimate 3.2k.

Which raises another question -- does Red allow you to interpret the footage Canon-style? As in, don't demosaic it at all, but instead just use each 2x2 block as the basis for a truly-sampled R/G/B pixel? That'd be an interesting option to have...
 
Razz16mm said:
Your GH2 test is a fair comparison as it removes unknown variables between different cameras for AVC 4:2:0 anyway.
Click to expand...
Okay, file is updated. It now has the 4K still downconverted to 4:4:4 1080p, the 1920x1080 raw, and a 1920x1080 AVCHD file all layered together. Unfortunately they weren't all shot in the exact same lighting conditions, but ... well, for purposes of color resolution and detail resolution it should still suffice.
http://www.mediafire.com/?4qb7k56edbzdf1d

PS: Who makes the wringer chart? I could use one of those.
Click to expand...
DSC Labs. It's one of my favorite charts now, I use it all the time. They have a revised version that's even a little more helpful (better-placed white patches, and res indicators on the circles) so it's even better than the one I have.
 
Barry_Green said:
A full frame downscaled version -- meaning, what -- non-etc mode? Or a still that's been downconverted?
Click to expand...

Same drill you did with the raw files. Shoot a full frame version encoded to 1080 AVC in camera and an ETC version encoded to 1080 AVC in camera. Be interesting to see how the color resolution holds up compared to the raw to RGB versions
 
Not sure how truly useful that would be coming from the GH2. It has to use some sort of binning or line skipping to reduce the data being read off the chip. ETC mode is a valid test because it is a true 1920x1090 raw Bayer scan, but that's not what the full frame gives. I tested the large frame in non-video mode because in stills mode it can deliver a full scan of the chip (in raw even) but it doesn't do that in video.
 
it would be useful to see how much info is lost with the GH2's video downsampling method, with respect to the full RAW downsampling that you already did - my guess is that it would be A LOT smaller loss than with the canons (and that includes the 5D3)
 
Can anybody tell me about the D.O.F from this sensor it is smaller than the regular dslr sensor so what is the influence ??
 
there IS a processor, just not a very powerful one - I have seen no mention of in-camera playback anywhere, but given that the files are stored uncompressed, reading them is basically the same as reading from the sensor...
 
Samuel H said:
there IS a processor, just not a very powerful one - I have seen no mention of in-camera playback anywhere, but given that the files are stored uncompressed, reading them is basically the same as reading from the sensor...
Click to expand...
From what I understand, there is no in-camera playback. You'd have to plug the memory card into a computer and try to play back there.
 
You must log in or register to reply here.
Back
Top