Digital Cinema by a dilettante
It’s not a pretty time to be a digital cinema camera manufacturer. There’s a crisis in the economy, film and tv production are down, reality television is supplanting dramatic content at an alarming rate.
And then there are the sensor wars.
In one corner, we have Panavision. Panavision probably needs no introduction to even casual sentient beings, but their reputation has been cemented on good old fashioned film gear, and their success in the new world of digital cinema seems to have as much to do with inertia and laziness as it does with superior camera design. Panavision makes a CCD-based camera called the Genesis, based more or less on the same technology powering the Sony F35, and all recent video cameras.
In the other corner we have Red. I have at various times called Red a fraud, the iPhone of digital cameras, and the most useless object I would probably still buy if I had the money. Red’s camera is based on CMOS technology, the same thing that powers most consumer digital still cameras and, apparently, most spy satellites and Mars landers.
Other contenders include Arri, most notably, as well as Thomson’s Viper and the aforementioned Sony F35 and F24 CineAlta cameras, and Dalsa, who have recently exited the digital cinema market and have stripped their website of everything cinema-related except for this article entitled CCD vs. CMOS.
Now, I’m not an image scientist, and I don’t really have a dog in this fight. I will admit up front that the arguments of those who back CCDs (Panavision, Sony) make more sense to me than those that back CMOS. I will also admit that I would not be surprised to find out that I am not only wrong, but an idiot. It would not be the first time.
CMOS cameras (Red, Arri, Canon, Nikon) market themselves primarily as high-resolution machines. Red’s literature in particular screams increasingly large numbers followed by ‘k’, seemingly placing pixel count above any other consideration. They achieve their resolution at film-friendly speeds by using a Bayer pattern, spacing an irregular number of red green and blue photosites in a pattern and then assembling an image out of that data further down the line. I italicized assembling because the war has gotten so prickly that I had to sit for a little bit to think of a word to use there that wasn’t “reconstructing” “interpolating” or some other loaded verb. People fly off the handle at such terms, as you’ll soon see.
This assembling is the chief weapon against CMOS technology, as employed by the CCD crowd. CCDs use an identical number of RGB photosites and thus (they claim) do not have to assemble anything. John Galt, creator of Panavision’s Genesis and the “Star Wars” F900, has an interesting and useful article detailing his perceived shortcomings of CMOS technology and disparaging Red and Dalsa. He throws a lot of chaff up because he hates the Dalsa people and, I assume, is quite threatened by the Red people, but his description of CMOS and Bayer from a technical perspective is fairly, um, illuminating. Also of interest is a periodically updated post called Red Facts that takes aim at Red’s megapixel count, color fidelity, and general worthiness compared to Panavision and Arri.
I am pretty amazed at the Red camera. It produces very large attractive images and it is downright cheap even compared to a Viper, which is basically a video camera. However, pulling keys from material shot on green or (heaven forbid) blue screen with a Red camera is unpleasant at best. You can get very satisfactory results pulling keys in Final Cut, which leads people to say lots of things that they can’t possibly prove, but at full resolution and with the goal of it looking good, things get dicey. Quadruply so for the Dalsa camera, which billed itself as an uncompressed, no-compromise 4k camera. I attempted to pull a key from green screen material shot BY THE DALSA PEOPLE, and was confronted with weird edges, random blocks, and things that reminded me of the sort of compression you’d see on HDCAM or YouTube. I might be willing to accept that I was doing it wrong, but as pulling keys properly is basically my job, the argument for my failure in this matter would need to be pretty compelling. These tests, brief as they were, did not make me like either the Red or the Dalsa cameras, and have colored (hee hee) my perception of CMOS cameras in general ever since.
CCDs are not without weirdness of their own, and if you take John Galt at his word, then the weirdness is a lot harder to explain. The original Genesis cameras were fraught with problems. In Superman Returns Adam Sandler’s Click, one scene featured backlit miniblinds separated by 2-inch dividers (sills? what are those called?). The light streamed across the dividers, creating the illusion that there was one large window. This was not a flare, or a lens anomaly, this was a sensor error. Apparently these kinds of things have been ironed out, but still, without the excuse of interpolation or, ahem, assembling, I’m left to wonder what they’re doing in there that would cause something like that. Additionally, Star Wars looked like absolute shit.
Nevertheless, Panavision’s Genesis and its sister the F35 have become pretty popular for filmmakers. I see them constantly in New York now, particularly on commercial shoots. The workflow is pretty straightforward, Sony now makes a deck that can accept HQ HDCAM-SR tapes, allowing people to shoot with very low compression to a tape format, which is still a lot easier to deal with than piles of hard drives and millions of dpx files. The Panalog LUT system is well documented and hasn’t really changed at all since its inception, so there’s a great deal of confidence there. Its easy and it works, two things that make them very popular with people like me.
By contrast, the Red post workflow is a screaming clusterfuck of massive proportions. Rather than decode the Bayer data on the camera, which would be practically impossible, Red leaves it to post production to deal with translating the raw image data into visible, usable pictures, a process that happens at about 3 frames per second on a good day. It is a testament to the marketing prowess of Red that they have invented a camera that requires such effort to get images from the camera to the screen, but that people are still lining up to use.
This will be fixed, eventually. Or at least dealt with. As a result of its popularity, or its buzz, equipment vendors like Avid (the editing giant) Autodesk (of Flame fame, the funnel through which 90% of all commercials are pushed) and Baselight (the hippest color corrector on the market) have been racing to integrate Red decoding into their software, so the problematic little r3d files spit out by the Red cameras can be read in without having to go through that annoying and messy 3fps transfer process.
My question is, why? The Red camera is an achievement, but is it better than the Genesis? Is it even better than its CMOS cousin, the Arri D21? I do not actually know.
Like I said at the outset, the pro-CCD, anti-CMOS arguments make logical sense to me, and my tests have only reinforced my confidence in non-Bayer images. It was this inclination that led me to send an article about a Dalsa/Panavision/Arri shootout in 2005, along with a catty little message, to a friend who knows Alan from Dalsa (who is no longer there), and would, I thought, be amused by what sounded like a very bad day for him.
Subject: dalsa and panavision
I’m casually looking into digital cameras at the request of my superiors, and ran across this article that seems to be having a bit of fun with your pal Alan:
http://www.showreel.org/memberarea/article.php?32
“the algorithms will take care of it” appears to be his new mantra.
-kirk
My friend then forwarded my email to Alan.
For being so annoying, I was rewarded by an email back from Alan, via my friend, which, after some unkind words aimed at me, went like this (before you even try, I’ll warn you that the Dalsa links are probably dead):
More to the point Geoff Boyle is a fuck-pig. Having said that look at Boyle the fuck-pig’s tests here that proved our point:
http://www.cinematography.net/digital-latitude.htm
then take a look at our own disclosure which takes it a step further:
http://www.dalsa.com/dc/origin/latitudetest.asp
then we go even further:
http://www.dalsa.com/dc/origin/datamapping.asp
Nonetheless lets look at the facts:
1. The ZODIAC movie, was shot on a Viper. It might be beneficial to understand that DALSA makes the fucking CCD sensor in the VIPER. We consider it a “B-grade” video CCD, nothing more. Not to say great images cannot come from the camera; indeed they can. However even in “Filmstream” mode you top out at about 9+ stops of exposure latitude and VIPER shoots ONLY at 1920×1080. Great if you are doing video, but not the schiz for features. Oh, and did I mention that WE MAKE THE FUCKING SENSOR?
2. There is no compression whatsoever coming out of our camera. None. They are 16 bit linear uncorrected HDRI DPX files in RAW Bayer format. I do not understand what “error correction algorithms” you are referring to. Hopefully not anything Boyle the fuck-pig said, as he does not understand even the basics of what was told to him. The camera has a great deal of ‘correction’ going on internally: tap balancing, crosstalk correction, etc., but nothing Kirk would have to worry his pretty little head about. It is all done in real time in the FPGA.
3. “Are you ‘reconstructing’ part of the image data ‘lost’ in the bayer pattern?” you ask…I say shame on you! SHAME I say! Exactly what image data is “lost” in the Bayer pattern? This is one of the great fucking myths about CCD’s, propagated by sniveling, ignorant little fucks on the internet. Look, I do not for one minute propose to be an expert in CCD design; on the other hand I have actually met experts in CCD design, including the guy who actually wrote the book on CCD’s: Albert Theuwissen, who coincidentally works for DALSA.
http://www.amazon.com/gp/product/0792334566/sr=1-1/qid=1138926718/ref=sr_1_1/002-4256071-9522452?%5Fencoding=UTF8
Also I actually have had to study CCD architecture very closely in order to understand how our camera functions. Navigate to this page:
http://www.dalsa.com/dc/documents/documents.asp
and read the document:
Technical Paper on Image Resolution of Camera System
What it basically tells you is this: the way we do Bayer sampling DOES NOT LIMIT EACH PHOTORECEPTOR SITE TO ONLY THE FILTERED COLOR! All of the photoreceptor sites capture all the spectra of light falling on the image plane, they are just BIASED toward a certain
frequency based on the Bayer mask. We make the goddamn CCD, the microfilters, and we WRITE THE GODDAM IMAGING
COEFFICEINTS!!!!! Don’t people think we might have done our motherfucking homework? If anyone wants to come in and test this, tell them to have the fuck at it. Bring color charts, High MTF samples, anything that can be thrown at the test. This is the single most misunderstood point in all CCD imaging: color reconstruction is not interpolation!
I did not go to MIT. I am not smarter than Alan, or Jim Jannard, or even Geoff Boyle. But this sounds like semantics, and the proof is in the images. Red images, properly processed, look great but are at best difficult to work with in a vfx environment. This is because they are compressed and have undergone some level of ‘color reconstruction’ or interpolation or assembling or whatever. Dalsa images, even when shot by Alan himself, were very difficult to key properly, and a lot of discussion at the time centered around the basic logic of looking at a file browser containing a directory of Dalsa frames and saying “the images have to be good, look how BIG they are!”
I do not care, ultimately, what camera people use as long as the images are acquired properly. I would prefer people continue to shoot film, and I would prefer that film be scanned at as high a bit depth as possible through the best lens available. This will not continue, however, and people will continue to ask me, as a post production professional, which camera I think is best. At the moment, I would tend to say the Genesis, because I and my colleagues have had the most luck with Genesis images. Further testing may swing things toward Red, who knows.
One thing will always remain, however, and that is that I will always enjoy annoying Alan Lasky.
[update]
As if by magic, look what showed up on todays shoot:
Leave a Reply