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

@Barry_Green
there is some problems ...
dbolex use some internal processing to record data from sensor to dng + internal ssd for buffer...
second how can u be so sure that everything will work they didn't make one prototype ...
its ok to say i got idea and i test some parts ...
but to say i got prototype and its shooting already send me a money to start production and u have nothing ... :(
sorry but that is not how u should run business
 
I really think Dan Hudgins should be involved with this camera - I think he can really help in regards to the DNG processing - the man is a genius! Hope they get in contact with him!
 
S. Matthews said:
whole thread about this on reduser and if I remeber correctly even on this site where he has disagreed with you on that...
Click to expand...
I've read all through what he's written, and I don't think we disagree at all. I think I've perhaps stated it blunter than he was prepared to, but I would be really surprised if he actually disagreed with me over the basic premise, which is: a 2K image from a Bayer sensor will not have more than 1K of red nor more than 1K of blue in it, which is what people think of when they think of 4:2:0.

"Neither - it's RAW. 4:2:2 etc. refers to the situation where chroma bandwidth is reduced in a Y'CbCr encoded signal to achieve a smaller bandwidth.
Click to expand...
This is entirely correct, and yet it doesn't mean what people want it to mean. He's saying that they have not reduced the color sampling down to something like 4:2:0, because that's a video processing task and their footage is not processed at all, it's raw.

Yet just because it's raw, doesn't mean that it's 4:4:4! And I think that's what people want to try to take from it. And the fact remains, there is not enough color sampling coming off the chip to make 4:4:4, no matter how you slice it.

Let's put it like this: a three-chip system delivers true 4:4:4 to its image processor. That image processor then converts and subsamples that data into whatever the desired Y'Cb'Cr signal is -- typically 4:2:2, but it could be 4:2:0 for recording to a 4:2:0 recording format. Fine. That's correct. And Graeme said that they don't throw anything away. Fine, that's correct too. But the difference is: the three-chip system actually delivered a true 4:4:4's worth of image to start with. The Bayer sensor does not. The Bayer sensor starts out with 1/4 the red, 1/2 the green, and 1/4 the blue. The three-chip system had 100% red, 100% green, and 100% blue to start with. The Bayer system has 25% red, 50% green, and 25% blue.

You can't make 100% red, 100% green, and 100% blue out of 25% red, 50% green, and 25% blue. You can't.


"...please don't use imprecise terminology. 4:2:2 etc. don't apply to a measure of chroma resolution in any sense, but a process by which resolution is deliberately reduced by some circuitry or process for bandwidth reduction reasons.
Click to expand...
See, this is again correct, I have no problem with this statement. I only have a problem with how people interpret what the statement means. See what I said above -- the 4:4:4 image from the three chips is reduced down to 4:2:2 or 4:2:0 (or, in the case of DV, 4:1:1, or in the case of HDCAM, 3:1:1). We are in complete agreement as to what this means.

Where we differ is in how people choose to interpret what this statement means. People want to believe that "hey, they're not throwing anything away, so therefore it must mean that they're retaining 4:4:4's worth of data!" And that's not true. It is true that they're retaining every ounce of what the sensor imaged, but what the sensor imaged is 50% green, 25% blue, 25% red. The sensor doesn't image 100% green, 100% blue, and 100% red, and that's what you'd need to make a true 4:4:4 image. And to make that, you absolutely must have three chips (either using a prism, or sandwiched together like a Foveon). Or, alternately, you need to downscale down to the resolution of your least-sampled color (which is what the 5760-pixel-wide Panavision sensor does, it downscales to 1920).

For instance, you could record a VHS tape onto Digibeta at 4:2:2. At which point you definitely have a 4:2:2 signal, but that actually tells you nothing about the measured chroma resolution of the signal. Similarly a BetaSP dub to the same Digibeta would be 4:2:2 but the actual measured chroma resolution would be vastly superior. A direct camera recording to that Digibeta would also be 4:2:2 and depending on the camera could have a vastly superior measured chroma resolution.
Click to expand...
Agree 100%. And this actually makes my point completely. He says if you took a VHS tape and recorded it using a 4:2:2 format, the signal would be 4:2:2. But the underlying data would not measure up to what our understanding of 4:2:2 would be whatsoever! You'd have a 4:2:2 recorded image with about 4:2:0's worth of color information in it. If you took a BetaSP dub (which has roughly equivalent 3:1:1 worth of color) and recorded it at 4:2:2, you'd have a 4:2:2 recording that contained about 3:1:1's worth of color. And if you took a direct camera recording off of the DigiBeta's three chips, you'd have a 4:4:4 source that gets chroma-chopped down to 4:2:2 and recorded as 4:2:2, and it'd have vastly superior measured chroma resolution. I agree completely. And that's what I'm saying about the Bayer system -- if you call it "4:4:4", what you'll end up with is the equivalent of that BetaSP dub into 4:2:2 -- you'll have a 4:2:0's worth of image being recorded "as" 4:4:4. But that doesn't make it 4:4:4.

Graeme may have disagreed with me on my less-technically-precise use of the terminology, but I sincerely don't think he was disagreeing with me on the fundamental validity of the notion that what a single-bayer-chip sensor produces cannot be classified as 4:4:4.


The issue of how colour filter array sensors see colour resolution compared to luma resolution is a different issue. In such a system we're dealing with broad colour filters where any detail is visible in all three channels in most circumstances
Click to expand...
Again, I agree 100%, and nothing I've written should be construed as me disagreeing with him, or him disagreeing with me. What he's referencing here is the basic "magic" by which that 50% green can then be transformed through the de-bayer process into an image that results in 80% detail. By using the broad color filter and borrowing detail information from the surrounding color pixels, a sophisticated algorithm can result in a synthesization off the sampled image that approaches about 70% to 80% of the nominal resolution of the original image. We agree.

But that's luminance. Not color. You don't pick up red color information out of the blue pixels. Color information is not synthesized accurately from an alternative color pixel's image.

So let me restate it in a more (hopefully) understandable way -- the marketing page that Red uses where it claims 4:4:4 is what both Graeme and I would object to, because a raw encoding is not a 4:4:4 encoding! 4:4:4 means an encoded, processed video or RGB image where there is one color sample of red, one color sample of green, and one color sample of blue for every pixel. And that is not what a raw bayer pattern delivers. And that is not what Redcode encodes. Redcode encodes the raw image, not a processed image. So neither Red, nor DBolex, can be claimed to be 4:4:4. Nor, accurately, can either of them be claimed to be 4:2:2 or 4:2:0. They are, inherently, raw systems that don't have 4:x:x encoding applied to them.

There. I said it technically accurately. And now people will ask "what does it all mean?" And at that point I'd say "well, it means that the images coming off this camera, after being processed, will have about 75% as much horizontal resolution as it has pixels, with half as much horizontal color information as it has luminance, and half as much vertical color information as it has luminance. You can think of it as 4:2:0 if you want, but that would be an imprecise shorthand to describe what's really going on."
 
vanvideo said:
So...why not just buy the GX2300?
Click to expand...
Well, perhaps a number of reasons. I googled it and for one thing, the GX2300 I found is using the monochrome version of the Kodak sensor, so that's one big problem. Second, it doesn't have a viewfinder or a normal video output, so it would be quite a bit more difficult to actually work with. And it's not shaped in a film-friendly bolex form factor. It doesn't record anything to CF or SSD or anything, instead it's a chip that spews its data out a Gigabit Ethernet connection.

What the DBolex guys are doing is repackaging that sensor in a film-friendly package that makes it usable for digital cinema usage.
 
vladszp said:
dbolex use some internal processing to record data from sensor to dng + internal ssd for buffer...
Click to expand...
Agreed they will be doing a little work. No image processing, but they will be formatting the files for DNG recording. They also have to have some sort of rudimentary video processing on it to make the 320x240 viewfinder image and the 640x480 b&w "video tap" feed.

second how can u be so sure that everything will work they didn't make one prototype ...
Click to expand...
You do have a good point as to that they must have a prototype before they know it will truly fully work. However, who's to say that at this point in development they don't have some manner of prototype? Maybe they shot original proof of concept footage before moving to the prototype stage? I don't know.

but to say i got prototype and its shooting already send me a money to start production and u have nothing ... :(
sorry but that is not how u should run business
Click to expand...
You do make good points here and it is something that any potential buyer should carefully scrutinize before putting their money down. I'm confident that the tasks ahead of them should be relatively simple, considering just how simplistic their total design is. It seems to me their proposed delivery schedule is quite aggressive. Who knows though, because the design is so incredibly simple (and doesn't feature such things as SDI and full HDMI) it might be possible for them to easily make their schedule.

We have to remember this is a kickstarter, not an established business, so when you say it's no way to run a business, I can see that point, but I do also make some leeway in that they're up-front with us that it's a bootstrap startup.
 
vanvide@
no its shoots color too...
its used to shoot this
https://vimeo.com/38286282 :)
but is machine vision camera not film-friendly...
about dbolex i am very careful ...
i was ready to send them money but for now i wait ...
and dont forget ikonoskop spend 6 years working on a prototype ....
 
all I am saying is using 4:2:0 can be misleading to most people who don't understand that, yes you are correct in that you agree more than disagree(actually I think the only thing he would disagree on is the use of 4:2:0), but that terminology is not perfect for describing Bayer sensors accurately.

Honestly I think I understand it but can't express it, all I know is I have much more room in post to work with my footage than with my old gear... that's all that matters to me, terminology and numbers are one thing, real world results are a whole other thing. I guess in essence I trust in Graeme's sampling theory to reproduce the best possible range of information from the sensor that he can get at this point with the system and resources he's working with, without sacrificing in other areas to achieve it.

so as you say at the end it's imprecise terminology so I feel it shouldn't be used in this instance, but hey I'm still learning all this stuff so forgive me for my errors.

Man this stuff makes my head spin sometimes!

Have a good one, and always love discussions with you Barry, just wish I had more knowledge to actually take part properly...

P.s. at the end of the discussion they get into replicating three chip systems on one chip and where the future of sensors may go(there are some interesting ones coming out even now) and it's interesting, there is the whole argument that bayer see's color the same as human vision, but it would be nice to have a full actual (or as close to as possible) 4:4:4 signal to work with in post.
 
Last edited:
Simon Shasha said:
I really think Dan Hudgins should be involved with this camera - I think he can really help in regards to the DNG processing - the man is a genius! Hope they get in contact with him!
Click to expand...
Sounds like an awfully good idea to me! Of course, that would only work if he's not under some sort of exclusive agreement with the other company, but maybe that's the case. Let's someone tell Dan to offer his services to the Digital Bolex guys!
 
Barry_Green said:
Sounds like an awfully good idea to me! Of course, that would only work if he's not under some sort of exclusive agreement with the other company, but maybe that's the case. Let's someone tell Dan to offer his services to the Digital Bolex guys!
Click to expand...

Same! Hopefully he is not locked in to any contracts and sees the suggestion and jumps on board!
 
Other thoughts -- I hope I'm not making too many heads spin, I'm just trying to explain things as I see 'em and keep people's expectations realistic. Even if it ends up that it's no more than 1440x810 with a 4:2:0 image, heck, it's still 12-bit raw (which means no compression artifacts and no gradient banding) and it's still a goodly-sized sensor. Way bigger than a GH2 in ETC mode, if a lot smaller than a GH2 or AF100 or Super35 or APS-C sensor. It's still a goodly size, and with a really really fast lens (like the Voigtlanders) you could get acceptably shallow cinematic DOF with it. And it's raw, which should mean lots of latitude to push and pull and grade in post.

The downside? The camera will do NO work for you, folks. You have to do everything that you're used to the camera doing for you. Raw images are ugly images. Flat, unprocessed, soft... look back on some of Jim Jannard's threads where he tried to explain to people that you must never look at raw footage, because it's not made to be looked at. It's made to be graded. You should only look at footage after you've processed it -- which means you have to run a de-noise process on the footage yourself. You have to apply gamma curves yourself. If you want smooth highlight rolloffs, you'll have to apply a color grade that rolls highlights off smoothly yourself, because the chip doesn't do anything about that. If you want daylight color temperature, you'll have to color-balance the footage (not the camera, not the sensor, but the footage) yourself. If you want to shoot footage under tungsten light, you'll have to color-balance the footage yourself. If you want the footage sharper, you'll have to apply edge-enhancement or an unsharp mask yourself. If you want compressed footage, you'll have to compress it yourself.

If you want to shoot sports or news for the local TV station, you definitely don't want this camera. If you want to shoot weddings and events, and turn around a same-day edit, you couldn't possibly (reasonably) do it with a raw camera.

But if you want to shoot raw images and grade the daylights out of it in post, and you have the time and budget resources to allow for such a grading process, then raw's your friend. I can't believe we haven't seen Perrone Ford in here yet, seems like this would be exactly the type of camera he's always been looking for.

I wouldn't replace my "A" video camera with something like this. But for the right project, it looks like just what the doctor ordered, at a great price. If someone is shooting cinema and wants to shoot just like we always used to shoot film in the old days, and is willing to budget the time and money for the post-production grading process, the DBolex (on paper) looks like it'll fill that niche with no competition at its price point.

And, as always, buyer beware and caveat emptor. This is apparently an idea, not a product, and those of us foolish enough to commit money to it are not buying a product, we're hopefully financing a company that will in turn deliver us a product, but we all understand the "hopefully" part. It's a gamble. Don't gamble if you can't afford to lose the dough.

Okay, last thing -- on the SDI output, man, that's a head-scratcher. I've been scratching my noggin about that one, and here's my current theory -- I doubt (sincerely, a lot) that the DBolex would ever offer what we all come to think of as a true, real SDI output -- meaning, full-resolution, full-color, broadcast-caliber, the whole nine yards. I rather suspect that what the SDI option would mean is that it would be an alternative output for the black and white "video tap" that the DBolex already has. Which is nice, it'd let you plug the camera directly into your SDI monitor instead of whatever oddball 1/8" output it already has (which, I suspect, is the same type of connector that palmcorders use, probably a 1/8" minijack-to-RCA for a composite video output). So I would guess that the SDI option would be an SDI connector instead, but that the video signal (640x480 black & white) would probably still be the same.

For those who don't know what a video tap is, well, basically, when you're shooting film, there is no output to any monitor. Film cameras (at least, old film cameras, of the vintage that the Bolex was) had no monitor outputs, and by and large monitors didn't really exist back then anyway. So what you had was a flickery viewfinder. Period. And the DBolex seems to be simulating that by using a 320x240 viewfinder. Later, camera manufacturers started offering what we call a "video tap" and what that was, was a video camera photographing the ground glass focusing screen inside the movie camera. The camera's lens shows an image on a piece of ground glass, and that's what the viewfinder looks at, and the video tap was a way to photograph that ground glass and output the video tap camera's signal out to a video monitor so that other people on set could see what the camera operator was seeing. A video tap is a fairly terrible video output, it's usually black and white and low-res and it flickers as the movie camera's shutter rotates and flickers, but hey -- at least it was an image!

That's what I expect the DBolex guys have in mind for their video output -- the functional equivalent of a video tap. Hence why it's standard-def and black and white, because -- that's what a video tap was. I don't think they'll make it flicker like a movie camera did, that'd be silly...
 
Barry_Green said:
Well, perhaps a number of reasons. I googled it and for one thing, the GX2300 I found is using the monochrome version of the Kodak sensor, so that's one big problem. Second, it doesn't have a viewfinder or a normal video output, so it would be quite a bit more difficult to actually work with. And it's not shaped in a film-friendly bolex form factor. It doesn't record anything to CF or SSD or anything, instead it's a chip that spews its data out a Gigabit Ethernet connection.

What the DBolex guys are doing is repackaging that sensor in a film-friendly package that makes it usable for digital cinema usage.
Click to expand...

Yeah, I was being a little facetious. :)

About the b&w video tap - since the images will be RAW, and you can manipuate the image in post, including white balance, or whatever, is it necessary to have a color monitor? All you're looking for is composition of the shot, and lighting. Is this correct?
 
Simon Shasha said:
Barry_Green said:
Sounds like an awfully good idea to me! Of course, that would only work if he's not under some sort of exclusive agreement with the other company, but maybe that's the case. Let's someone tell Dan to offer his services to the Digital Bolex guys!
Click to expand...
Same! Hopefully he is not locked in to any contracts and sees the suggestion and jumps on board!
Click to expand...
In this post on EOSHD Dan Hudgins says that he's worked with several such companies, and from that I infer that he isn't locked into an exclusive arrangement with any of them.
 
Last edited:
Barry_Green said:
Understand this: an image sensor is a MONOCHROME device. If you see something advertising a single chip at 2048x1152, such as this Digital Bolex, then you know that that is the resolution only in black and white. Period. If you want color out of a single sensor, you have to put on some sort of Color Filter Array (CFA) such as (typically) a Bayer pattern. And when you put on a Bayer pattern, you get a colored filter over every pixel. And 1/2 of those pixels will be covered with a green filter, and 1/4 will be red, and 1/4 will be blue.

And that, right there, is why it is impossible to get 4:4:4 color at the stated resolution -- because only 1/4 of the pixels are blue, and only 1/4 of them are red.


Here's the bottom line: 4:2:0 means that you have an image that has twice as many luminance samples as it has color samples, on both the horizontal and vertical directions. And a Bayer-pattern sensor has twice as many green pixels as it has red or blue pixels, on both the horizontal and vertical directions. Therefore, it could be argued (and is argued by Panavision) that a Bayer sensor is a 4:2:0 sensor. The terminology can be argued the other way, that the sensor itself isn't any manner of x:x:x notation, and that argument would be technically correct while disingenuously misleading, because it's getting to the point where we'd be arguing over what the meaning of "is" is.

Graeme Nattress wrote up an excellent lengthy reply to this basic issue years ago about the Red, and he said that it's basically a little better than 4:2:0, because while there are only 2K's worth of color samples in a Red's 4k image, the loss of luminance resolution from the de-Bayer ends up making the color proportion slightly higher. So he argued that a Red's 4K image would be more like a "3.2:1.6:1.6" if you wanted to call it that, but that would be fairly meaningless in common terminology.

So let's just call it what it is: the images coming off a Bayer sensor are basically 4:2:0, for all reasonable intents and purposes.
Click to expand...

This is an overly simplistic view of Bayer color filter arrays. The color filters are not pure blue, red, and green like a 3 chip sensor array. Blue passes a percentage of all three primaries with a blue bias, same with red and green filters. This is why you can get a full resolution black and white image, which by definition contains all colors of light, from a Bayer color filter array sensor with bias compensation. Is color resolution lower for Red and Blue? Sure. But it does not in any way equate with any particular RGB or YUV color image encoding scheme all of which other than 4:4:4 discard color information from the sensor. A 4:4:4 encoded video file from a Bayer sensor will have much higher color resolution than a 4:2:0 encoded video file. A Bayer sensor can at best match the relative red / blue color resolution of human vision, which is comparably lower than green too. It is still the closest and most efficient mechanical mimic for how human vision samples resolution and color which is why Bayer sensors are so universally popular in all kinds of cameras.
 
Last edited:
Okay, another thing -- I've just spent a lot of time writing a lot of stuff which could have been saved just by pointing to Stu's blog entry. It's been listed before but here it is again:
http://prolost.com/blog/2012/3/13/digital-bolex.html

I pretty much agree with everything he's said there, other than about Red. Red bit off a much, much bigger chunk of the apple than the DBolex guys are trying to do. The Scarlet was supposed to be a "video camera" as well as digital cine, and at higher res, and the customers they got were not the customers they could deal with. The key to the success of the DBolex will be (in my opinion) managing expectations. When the Scarlet was first announced, people really seemed to think it would be the be-all, end-all, do-everything, every-other-manufacturer-goes-out-of-business camera. And of course it could never be that. Well, neither will the DBolex. It will be a very simplistic tool, with the design goal being to get you the simplest, cheapest path to raw images.

Okay, you all knew it was coming, but it's time for a car comparison. Think of the AF100 in terms of a car, it's a, well, maybe it's a Honda Civic LX. It's comfortable, it's smooth, it's quick, it has power locks and power windows and air conditioning and a great stereo, and when you want to go somewhere you turn the key and go. By comparison, the DBolex is a 300HP engine on a go-kart frame. When you want to use it, you go blasting down the dragstrip, and then you spend the next week repairing everything that broke off it. (that's the equivalent to all the post-production you'll have to do on the raw images).

In other words, the AF100 is the daily driver, and the DBolex is the weekend hobby. There's definitely a place for both in a shooter's arsenal, as long as they can afford both. That's my take, anyway, and I do hope these guys can deliver what they set out to deliver.
 
S. Matthews said:
Have a good one, and always love discussions with you Barry, just wish I had more knowledge to actually take part properly...
Click to expand...
Hey, no problem, I understand exactly where you're coming from. If you want to understand more, check out this post, I found it from the ProLost links and it really explains it pretty well...
http://magazine.creativecow.net/article/the-truth-about-2k-4k-the-future-of-pixels

there is the whole argument that bayer see's color the same as human vision, but it would be nice to have a full actual (or as close to as possible) 4:4:4 signal to work with in post.
Click to expand...
There is that argument, yes -- and that's another one that people have to be really careful that they don't get confused by what is being said there. Your instinct to prefer 4:4:4 is correct, we'd all love that if possible, and there's no rational person on earth who could argue that Bayer-pattern footage would be superior in any way to 4:4:4 footage.

Look at it this way... understand that 4:2:0 is a serious savings in data and that it results in an image that looks really rather good to our eyes -- that makes sense. But our eyes aren't the only place footage gets used, especially when you're talking about digital cinema and the compositing and post processes it will go through. I mean, it's simple -- show the average person a video image at 4:2:0 and one at 4:2:2, of a full color scene, and ask them to pick out which is which, and I think the vast majority of people would be unable to do so. But try compositing a greenscreen shot at 4:2:0 and one at 4:2:2 and any one of us would easily be able to pick out the difference. 4:2:0 is a good compromise for video footage, but for heavy work in post obviously 4:2:2 is better, and 4:4:4 would be even better than that.

Where people need to be careful is that they shouldn't go thinking that the argument is "bayer replicates how our eyes see." I mean, that's basically saying our eye has a Bayer pattern, right? And that's just not true by any stretch. Wikipedia says we have about 120 million rods, and only 5 million cones. That would make a ratio of about, well, what would it be... 4:0000001:0000001 or something? That's silly. Now, if you wanted to say "Bayer is the sensor pattern that most similarly reproduces the way our eyes see COLOR..." then yeah, that's a more fair statement. But - so what? It certainly doesn't replicate how our eyes see images, so -- what's the point of bringing it up? The camera sees images and delivers them to the computer and they get processed. Bayer is a fine way of doing it, a popular way, and has been around for 40 years and will be around for a lot longer. That's fine. I'm not complaining about Bayer at all. I'm just saying that it is what it is, and it isn't what it isn't. And I think it gets misrepresented sometimes.
 
You must log in or register to reply here.
Back
Top