Rai & Markus' "Drake" HD camera

[Markus Rupprecht]

Important moderator notice:

This thread originally was part of a much larger thread to build
our own HD camera's. For various reasons it has been split off
into its own thread. Please keep in mind that some of the earlier
conversations may read a bit strange due to this split (some
parts may exist in the old thread, etc.). You can find the old
thread in the following place:

http://www.dvinfo.net/conf/showthrea...threadid=25808

Thank you for your consideration, back to our regular broadcast:
==============================================


Hi everybody!

My name is Markus Rupprecht. I'm the producer/director to the fantasy movie project Rai refered to. The camera creation team agreed not to participate in these hundrets of "soon we will have this and that" threads but wanted to complete our development, get a prototype runing and do some serious work with it before spreading the news. So that is why he didn't tell a whole lot. On saturday and sunday we had a first time shooting, both indoor and outdoor scenes, day and night with the camera and are curently processing the material.

We build the camera to match film workflow, so sound is recorded seperatly on DAT, we don't have a sync connection. The camera produces a "beep" sound and records red frames like a movie camera does. Scince the technical informations are stored in the file header of the recorded file we don't use a slate. A small detail, but it did increase the workflow very nicly.

We use a CMOS sensor that is capable of producing very natural images, no video look at all. I'm stunned by the perormence. It's cinema. The price we have to pay is that the cmos sensor needs a lot of light. Comparable to 25 ASA film material.

The night scenes were shot with a 1:0,95 f-stop lens with open aparture. It's a hell of DOF. And a hell of focus pulling. But we managed complex dolly and crane moves.
But this is definitly nothing for unskilled hobby filmers.

We also developed a rather complex software, not only controlling the sensor and recording the data to hard drives. The software is capable of eliminating issues like fixed pattern noise and gain noise. It does that already almost perfect, what is left in terms of noise is changed to random noise, that looks like film grain, so not too bad.

Scince we record uncompressed it is possible to run several algoryhms over the material to enhance the quality and to adress certain cmos behaviours in certain light conditions. That is done in the "processing" phase that is currently running in the room beside the one I'm sitting in now.

But this is just technological stuff. What we now have is a instrument to tell stories. To have a indi budget and shoot cinema quality with all the benefits that digitally recorded material has. And the best is: it works.

Unfortunatly I will now leave you allone with those informations. Please be patient, soon we will release some images.

Greets

Markus Rupprecht

[Jason Rodriguez]

Hi Markus,

Is this camera shoulder-mountable, or is it thethered to another device (PC somewhere)?

What CMOS chip/framegrabber combo are you using?

How much lag do you have in your preview, and do you use a viewfinder, or again, is everything externalized on another PC on-set?

[Markus Rupprecht]

sync is indeed no problem in terms of "keeping sync" but you need a starting point. That's why you use a slate - to record a "starting information" on your recording devices.

As I said, we will post detailed informations soon so we won't answer all questions directly but yes, the camera is portable, it can be mounted to a tripod or steadycam or whatever, it can be oberated with batery. It has a 17" viewfinder and the software does allow realtime preview in full 720p resolution, so no lag.

We did timing critical pans and focus things and it all worked well. We don't use special grabber cards, just regular PC parts and buses. Well, we searched a while to find good components that are not power consuming and had to to even build some internal devices. I'm no engeneer, so Rai or somebody else from our crew will propably answer specific questions in the future. But let me tell you there won't be something like the "10 steps to build your own HD camera" tutorial. It's complex like hell.

We tried several industry camera heads. They are all good in theory. And when you have perfect daylight, life is beautiful. But it wehen it gets darker you get into trouble. So between throwing pieces together that can record 24 frames per second and turn the thing into something that can record 24 frames in cinema quality and that allows a film workflow did cost us more than half a year with five people, all pro's. When all the problems in the beginning showed up I decided not to do the self development of such a camera and instead to buy the new JVC 720p camcorder. Well, it sucked more than the problems we had to face. You know them, this forum ist full of problem descriptions.

So my advice to anybody trying to do something similar is: Better be serious in what you want and what you want to do with it. And that should be making great films, right?.

[Jason Rodriguez]

How long were your longest takes?

If you're not going more than 1 minute, you'll never notice any drift. It'll typically take a long time (at least 10-20 minutes) for drift to occur, depending on how accurate the camera's fps is.

BTW, how did you get by with no framegrabber? Doesn't the camera have to hook up to the computer somehow?

Also the ProCam 3560 seems to be rating at ISO320-400 at 0db gain. According to Altasens, ISO1600 is at 18db.

Still curious what camera head you guys are using.

[Markus Rupprecht]

For all german speakers among you: Under www.hackermovies.com you will find the first article of a series describing the development process of our HD cam. It's called "DRAKE". So if you don't speak german, go for the key word DRAKE, in the article you will find some early camera images, already done with a mobile prototype.

And to move a little away from all these boring technical discussions I'd like to introduce some beauty to the topic. The first processed shot from our premiere weekend.
It's divX compressed 720p, aprx. 8MB in size. Make shure you got enough screen resolution, watching it downscaled is no fun.
Here is the download link:

www.drachenfeder.com/int/take1.avi

It's shot with a 25mm 2/3" lens f-stop: 0,95, about 4 meters of dolly / crane move and 3 focus points. It took just 3 takes to get it perfect. So I guess the damn thing is usable. Enjoy the pictures as we do. We've build the indi viper!

[Wayne Morellini]

Here is a pretty poor english translation,:

http://translate.google.com/translat...language_tools

Print version:
http://translate.google.com/translat...language_tools

[Soeren Mueller]

Hi Markus!

Damn it's really a shame that I missed the last "DRAKE tech meeting" in september! :-(
Congratulations for finishing your first (test) shot! Looks really great.. however I still noticed (esp. in the background) this uniform pattern (is this still from the bayer filter?!). I already noticed it in the earlier frame grabs I've seen from DRAKE - afaik Steffen Hacker noticed this too. ;)

Nevertheless it sounds pretty darn cool and perhaps I can convince you one day to build one more :-))

Cheers,
-Sören

[Jason Rodriguez]

Hey, how are you getting such large F-stops? Are these custom-made lenses?

I do not know of any 2/3" lenses from Angenieux, Zeiss, Optima, Cooke, etc. that open up that wide.

I'm trying to make some sense of this whole thing here and there's no information about what camera head, how you've input into the computer, what lens you're using, etc.

Please don't blame me for being confused, again, I'm just trying to make some sense out of this whole thing, and the details still seem quite sparse.

I'd be very curious to know what the specs on your machine (digitizing computer) were.

Thanks for uploading the clips though, greatly appreciated :)

[Jason Rodriguez]

Hey Markus,

Just took a look at your movie there, it looks very nice, but I think I may have spotted one problem that I think may be giving you that "ASA 25" rating you're talking about.

Now I have no idea about the technical details concerning your camera because you haven't listed which camera head you're using, so there's no real way of telling. But Juan could probably back me up on this one just by watching the film.

Those clipped over-exposed highlights should not be happening. You're getting a very harsh clip which tells me that you may be using the top 8-bits of the image, or that this is a "10-bit" camera, but there's so much noise in the lower two bits that it's essentially an 8 or 7-bit camera (like the Micron chip we first experimented with here).

With the Altasens for instance, you can come away with a VERY dark image, that meaning that the white chip on a Macbeth is exposed at IRE 10, (yes, ten!), and still have a perfect image (there would be some noise of course, but it's equivalent to using a Canon DSLR like the D60-that's my point of reference-at ISO400 or 500, so nothing to worry about). That's because the Altasens for the 4095 values it encodes at 12-bit has greater than 11-bits of real-data (only the bottom 3 values are noise-that's from the head engineer over at Altasens).

So you may be only getting ASA 25, but it's not because the chip itself is insensitive to light, but you may not be capable of taking advantage of the lower bits due to noise and other artifacting or non-linearity (again, not knowing what the chip or camera head is, there is no way of telling).

BTW, how did you get around rolling shutter "skew"?

Another thing, that "uniform" pattern, that's not a CMOS issue, are you using a ground-glass for 35mm DOF simulation? The "uniform" pattern looks too out-of-focus to be fixed pattern noise.

[Soeren Mueller]

Hm are we talking about the same noise here? Maybe you have some post processing activated in your divx decoder filter ;-) ... I made these two captures from the avi and upsized them 200%:

http://www.soeren.de/misc/pattern1.png

http://www.soeren.de/misc/pattern2.png

This sort of "ordered dither" pattern is clearly visible in all bright but not near-100%-white areas of the video - and it is fixed and not moving! (can't be from some ground glass)

[Jason Rodriguez]

Okay, nope, you're right, that looks like something from a de-bayer filter, like some of the pixels didnt' get interpolated or something odd like that.

Actually on my CRT monitor here, I didn't see that stuff, but stuff like that can get lost on CRT's. Are you on an LCD?

Also are you guys using variable gradients, or some other form of bayer demosaicing?

[Markus Rupprecht]

I knew that everybody would jump on the fixed pattern noise thing. Oh, well. We know that it's there and we found a way to eliminate it. thing is that for every gamma curve you apply during shooting you have to define a huge set of paramters to get enough data so the software is able to filter the noise out. We have to do it ones for all time to come, but the calibration module is a fresh feature, so for the weekend we couldn't feed in all the data we would have needed for a perfect picture. I decided to use a gamma curve on this shot where the paramters were not perfectly balanced. Well, after all it was a test. I was prepared to kick the entire material into the garbage. It turned out better than expected. And after all the quality of a movie shot is not only defined by technical issues on pixel level. Please, all you engeneers out there, keep in mind that those machines are ought to produce art, not sets of clean signals. Anyway I uploaded one of the test clips with already a rather good FPN removal in place. You'll notice the difference.

Find it here: www.drachenfeder.com/int/dach_after.avi

We need a little more time and more and more test runs. But the methods are proofed to be allright.

The cispnes in the shot is the way we want it to look. The original camera picture has more information in the dark parts than it seems. So, just be patient. We do our work and when we have the perfect set of comaprison shots we post them.

Greets
Markus

[Soeren Mueller]

@Jason ... yes I'm lucky to have an Apple 23" CinemaDisplay which sports a res of 1920 x 1200 - very nice tool for HD monitoring. :-)

@Markus - sorry... it wasn't meant as a critique in any way.. i just wanted to know what I was seeing there.. just tried to understand, how your cam works and what type of filtering you have to apply... any chance to see some real "raw" (totally unprocessed) data from the cam? Ok sorry again.. guess we just have to be patient until you've got time to post more details. 8-)

[Soeren Mueller]

PS: The new vid looks much better pattern wise.. however on a high-res display (TFT) there is still some slight pattern visible!
However: the image looks really nice and much more filmlike than any other HD cam in that "price range" (if you can say that already... taking into amount all the work that went into the project)

[Richard Mellor]

thank you for the footage markus: question - are you using a35 mm adapter type device. with rai's help and original drawings we are working hard on the ocillating adapter prototype.