Digital Bolex weeks away

Darren Levine said:
Advancing tech always comes into play. imagine if the gh3 hack team manages to get raw? or the rx10, or the nex5/6/7? the only two tricks the db has that keeps it treading water are raw, and arguably the CCD sensor. i say arguably because you have to consider the vast amounts of high end cameras shooting brilliant CMOS images. rolling shutter on cmos cameras has been very well tamed on most new releases.

So ask yourself: if the DB shot only avchd 24mbps, how much of its appeal would it lose? do you really like the form factor? do you really like the feature set?

First gen is always the rockiest, and if the db team decide to go for a v2, i would bet on them putting out a respectable competitor, seeing as they do take in much of the input the users give them. trouble is, by the time they do that, blackmagic will be in v3 and have worked out their kinks. and the majors will have released dozens of new interesting cameras. fading into history is far too easy these days, and thats where RED has it spot on with their modularity/upgradability.
Click to expand...

Yes I do like the form factor, much better than any DSLR form factor for shooting motion, and easily rigged to just about any state you would want.
Is it a general purpose video camera? No. It sacrifices consumer convenience features for IQ and sound quality potential that no AVC camera can match.
AVC is not relevant. It is IMO a poor codec for camera original acquisition as implemented in cameras in this price range and 4:2:0 color is not acceptable to me. The differemce between shooting AVC and the D16 is as great as the difference between AVC and film.
 
Razz16mm said:
Yes I do like the form factor, much better than any DSLR form factor for shooting motion, and easily rigged to just about any state you would want.
Is it a general purpose video camera? No. It sacrifices consumer convenience features for IQ and sound quality potential that no AVC camera can match.
AVC is not relevant. It is IMO a poor codec for camera original acquisition as implemented in cameras in this price range and 4:2:0 color is not acceptable to me. The differemce between shooting AVC and the D16 is as great as the difference between AVC and film.
Click to expand...

then you've effectively agreed to my point...
 
Razz16mm said:
A reasonably priced high quality replacement for 16mm film cameras is exactly what many of us have waited in vain for from other manufacturers.
Click to expand...
That I won't dispute.
The image has a very smooth organic film like quality free of moire and aliasing under conditions that would make BMPCC video unusable. With excellent lenses, resolution and detail are close to as good as HD gets. It looks fantastic on a cinema screen with a high end 3-chip DLP projector.
Click to expand...
To talk about it being free of moire and aliasing, I assume you've seen the results it gives on charts? Because for all the reasons I've laboured before, for a 1080/2K output, with a 16mm chip spec of 1920(2048)x1080, nobody should expect low aliasing.

Well, maybe with a very aggressive OLPF - but that would have to be such as to seriously impact resolution.

Whilst it may "look fantastic on a cinema screen", then the same could be said of many films in isolation - I cannot believe with the specifications it has that "resolution and detail are close to as good as HD gets". We've been through this before. To get to that level the laws of physics require something like 3 1920x1080 chips, or an oversampled single chip (as the F3 or Alexa). If we said that either of those cameras really were "close to as good as HD gets", then are you seriously saying that in a side by side comparison the DB would match them for resolution and detail?

If so, then please post zone plate chart results to prove it.

The limitations of single Bayer 1920x1080 sensors for HD are nothing new, the science is well established. As example, see Alan Roberts report on the XF105. ( http://downloads.bbc.co.uk/rd/pubs/whp/whp-pdf-files/WHP034-ADD65_Canon_XF100-105.pdf ) He certainly does not call it "bad", but it's worth reading his general comments and looking at the zone plates on pages 11-12:
There are strong diagonal aliases, and coloured horizontal and vertical aliases, which is a clear indication that the sensor has a Bayer pattern structure. Also, the fact that the coloured aliases are centred on the extremes of the pattern proves that the sensor is 1920x1080, confirming the specification in the manual.

In a camera with a single Bayer-patterned sensor, the red and blue patterns each have half vertical and horizontal resolution, because the pattern of pixels alternates between …RGRGRGRG… and …GBGBGBGB… line by line, thus red pixels appear in alternate columns and lines as do blue pixels. Therefore the clean resolution limits for red and blue are half that of the sensor.
................

The coloured aliases around 1920 horizontal and 1080 vertical are due to the reduced nature of the red and blue resolutions.

The only way to get good resolution from a Bayer-patterned sensor is to have more pixels, typically 2880x1620 would be enough to deliver full resolution at 1920x1080.
Click to expand...
Emphasis on the last paragraph is mine. Don't anyone expect the DB camera to have "resolution and detail {as} close to as good as HD gets". That's not possible. But if the Digital Bolex company have somehow achieved the impossible, then please, post a chart shot with it.
 
The practical upper limits of HD resolution if you want a smooth organic alias free film like image is around 800 lines vertical resolution and around 1.6k horizontal resolution. That is it max, regardless of what camera the video comes from. ARRI Alexa Prores HD output hits less than 10% MTF at 1.6k horizontal resolution, the effective upper limit of the camera and the format.

If you look at the test charts for any number of oversampling current gen HD cameras in the under $10k price range you will see they all top out at around 800 lines vertical. AF100, F3, FS100, FS700, C100, C300 all fall around that range. This level of resolution is within reach of state of the art debayer algorithms at 1920 x 1080 with no chroma moire and no aliasing that is not a result of the inherent limits of the format itself. Just because most single chip bayer video cameras are saddled with inferior debayer algorithms due to real time processing requirements and limited processing power doesn't mean that we have to suffer the same results from post processing raw files. State of the art fourier wavelet transform debayer algorithms pretty much eliminate chroma moire as an issue at all, but require 60 times or more the processing power of traditional methods. Definitely not possible in real time within the limits of what can be built into a low cost camera at the current state of the art.
 
Last edited:
Razz -- you have access to a DBolex, I presume? Can you just shoot a zone plate with it and settle the question once and for all?
 
Well, I don't get involved in this debate THAT much, because it's pointless without actual factual references. People argued to high heaven about the BMC, and I kept saying it was a terrible idea to ignore the OLPF, and they did it anyway, and the first time John Brawley posted a picture of a res chart that was all I needed to see to tell me that everything I'd suspected would be the case, was in fact the case. Stubbornly, and to the frustration of countless individuals, there still is no "magic". Things work exactly as they do and exactly as they should...

... so, in that context, I think debates about the DBolex's bayer performance should be tabled until they can be actually answered. The answer will be very easy to get, simply by pointing the camera at a proper res chart or (preferably) zone plate. And you know someone's going to do it, so it would behoove the DBolex team to be up front and just do it. There are cameras working now, they could just do it, so -- could you encourage them to, please?
 
I just don't see the pint in this camera anymore. They set out to be trailblazers and now they're just trailing... way behind.
 
Razz16mm said:
If you look at the test charts for any number of oversampling current gen HD cameras in the under $10k price range you will see they all top out at around 800 lines vertical. AF100, F3, FS100, FS700, C100, C300 all fall around that range.
Click to expand...
This is simply not true.

As example, Alan Roberts tests and zone plates of the C300 (C100 will be the same, same chip and processing) - http://cpn.canon-europe.com/files/n...report/Alan_Roberts_on_the_Canon_EOS_C300.pdf - bottom of page 12. This clearly shows resolution with minimal aliasing extending easily out to beyond 1000lpph.

Want an example with a Bayer filter and conventional deBayering? (And yes, in real time.) What about the charts Alan shot for the F3 - http://downloads.bbc.co.uk/rd/pubs/whp/whp-pdf-files/WHP034-ADD68_Sony_PMW-F3.pdf ? (Bottom of page 10, Figure 4.) Maybe not quite as good as the C300, but still showing real detail up to and possibly beyond 1000lpph. SOME chroma aliasing on high frequency vertical and horizontal lines - but nowhere near as much as for the XF105. That's the effect of oversampling beyond 1920x1080.

Whilst we're at it, let's see what was said about the EX1, just as an example of a 3 chip, 1920x1080 model. (I'd expect such as the XF305 or the HPX250 to be similar in terms of resolution.) http://downloads.bbc.co.uk/rd/pubs/whp/whp-pdf-files/WHP034-ADD30_rev2_Sony_PMW-EX1_and_EX3.pdf (p14, figure 3) Again, manages to get comfortably beyond 1000lpph.

And none of these show remotely anything like the colour aliasing on the vertical and horizontal axes, and the high level luminance aliases on the diagonals, that the XF105 does.

Just because most single chip bayer video cameras are saddled with inferior debayer algorithms due to real time processing requirements and limited processing power doesn't mean that we have to suffer the same results from post processing raw files. State of the art fourier wavelet transform debayer algorithms pretty much eliminate chroma moire ......
Click to expand...
The F3 also has to manage deBayering in real time - but manages vastly better results than the XF105. True it may have better algorithms, but the big advantage it has is oversampling. DeBayering has a lot in common with upconversion - some processes are better than others, but upconverting SD to HD can NEVER be as good as true HD. In the end all these processes are effectively guessing with the aid of maths.

Barry_Green said:
Well, I don't get involved in this debate THAT much, because it's pointless without actual factual references.
Click to expand...
OK, but the underlying maths and science is well known and generally accepted. True, we haven't got an actual example of results from a Digital Bolex, but knowing that it's a 1920x1080 Bayer sensor means it's possible to predict a huge amount about it's ceiling performance. (By which I mean a deBayering process that was efficient as theoretically possible.) Hence all the references above.
Barry_Green said:
... so, in that context, I think debates about the DBolex's bayer performance should be tabled until they can be actually answered. The answer will be very easy to get, simply by pointing the camera at a proper res chart or (preferably) zone plate. And you know someone's going to do it, so it would behoove the DBolex team to be up front and just do it. There are cameras working now, they could just do it, so -- could you encourage them to, please?
Click to expand...
And amen to that. I actually consider the Digital Bolex quite reasonable, overall. It's a niche product, but could do quite well in a niche market, and there's nothing wrong with that. But it's wrong to make it out to have a performance which is far better than should be expected from the specs - that's only likely to lead to mistrust. What theory would predict for res/aliasing performance isn't bad - there's no need to overegg it.

So I'll simply echo Barrys call. Can you encourage the Digital Bolex team to just publish zone plates from the camera, please?
 
Last edited:
Wow, the BMPCC just destroyed the 5D2 Raw... even though the 5D2 "looks" sharper, it's just aliased to hell and back. The actual fidelity of the BMPCC is quite a bit higher. Still got a ton of color aliasing all over the image, but vastly better than the 5D2.

You can still see it's got a long way to go when you compare it to the downrezzed NEX raw still, but as compared to the other video formats you listed the BMPCC is the clear winner.

Congrats on putting together possibly the most useful, most informative, most accessible camera test (that doesn't use charts) that I've seen. Even though I prefer the scientific clarity of the charts (which are purposely designed to reveal camera behavior) your test still manages to convey everything that I needed to know, and did so clearly, concisely, and extremely usefully. Well done!
 
It's very revealing Sam, and I'd agree, well done.

It's also worth noting that we should all be very careful about what is meant by "aliasing" - it's become something of a misused word.

Technically, it means something very specific, and you need specific tests to truly define an alias. Such all fundamentally rely on sloping lines, especially lines a few degrees off from the vertical or horizontal. Simplest is what seems to be a series of resolution wedges - but with the lines all sloping by a few degrees. True resolution gives an output with the slope the same as the chart. The moment it becomes an alias, the lines appear to slope the wrong way. Which is the advantage compared to a traditional chart with true hor and vert lines in the wedges - it's very easy to count aliased resolution as true in such a chart.

Resolution "trumpets" use the same principle. Move away from the center hor/vert lines and the slope increases. Result is that "true" resolution sees the lines converging (as the chart) - aliases represent as divergence. So the limit of resolution is the point where convergeance turns to divergence. And finally, zone plates are a way of showing the camera all possible resolutions and all possible angles simultaneously.

If you can't afford proper charts, there is something that can be tried very cheaply. On a patch of paper draw a series of parallel lines a few degrees angled to the horizontal. Then point a camera at the result and slowly widen the angle of view. If (looking left-right, say) the lines point down, then that's how the image should look. As the angle of view is widened (so the lines appear to get closer together) there should come a point when the lines will seem to point UP on the image. That's when you've gone past the true resolution limit of the camera into aliasing, and if you measure how far apart the lines are, how many are on the patch, how big the patch is in frame, and do some maths - you can work out the figure in lpph! Who needs expensive charts to be scientific! :)

Excuse the digression - I hope at least one or two may have found it useful! My apologies to those who are well aware of the above.
 
Thanks. You all should know: I'm blushing right now. :)

BTW: I do have a cheap ISO test chart that is good-enough for testing up to 4K, but for this test I really wanted to use a real-world target, since I anticipated that detail in highlights, midtones and shadows would be very different in some cameras, thus the dark textured backdrop that I guessed would make the uncompressed RAW cameras shine above the others.
 
Last edited:
Nice! I know I've been dying to see motion (both in frame and from the camera itself), and something shot indoors with natural light (rather than those hideous Kinos that the camera tests employed).

That's often (to my eye anyway) the difference between whether something registers as "cinematic" or "video" when all is said and done: how natural light registers, and motion. Specs, resolution, all that jazz are great and all, I just want to know if there's a way to get an evocative, cinematic image from the camera.
 
Shots of people.

See how shallow the focus can be. Maybe with a 20mm lens or thereabouts, wide open iris, medium shot (waist up) or medium close-up (head and shoulders). I know I could somehow get this from any number of Super16 films, but as long as you're shooting...

Dynamic range. A step chart would be interesting, but if not that then some bright outdoor scenes. However, I'm interested in what kind of image these raw cameras can get if people protect the highlights extensively, "underexposing" by even a stop or more than they normally would, so that in post you can make the highlight roll-off super-smooth. You would probably have to introduce a little bit of grain in the shadows, but I think I would actually like that. The film look most pleasing to my eye is that of some of the Oscar-nominated films from the 80s and 90s, which, unlike the surgically clean imagery of video today, had some grain.
 

Attachments

  • kramer_v_kramer.jpg
    kramer_v_kramer.jpg
    89.2 KB · Views: 0
Last edited:
You must log in or register to reply here.
Back
Top