4:4:4 10bit single CMOS HD project

[Ben Syverson]

Michael,

Yeah -- the idea is definitely similar to the Viper workflow, where you shoot RAW, and then have flexibility later.

720p @ 24fps in 10 bit == 26.3 MB/sec. Still possibly within the range of a 7200rpm laptop hard drive.

Also, keep in mind that with such a low data rate, you can simply capture to RAM. With 2gigs of ram, you could record about a full minute of footage in 720p 10bit. Then it doesn't matter how fast your hard drive is.

The amount of footage you can capture at 1080p in 10bit is much less -- about 30 seconds. Whether or not that's a deal-killer depends on your needs...

I'll be investigating bus-powered RAIDs -- I think it should be possible to bus-power a few mini Firewire800 HDs and raid them together to achieve 60MB/sec.

- ben

[Michael Pappas]

Hi Ben

<<<720p @ 24fps in 10 bit == 26.3 MB/sec. Still possibly within the range of a 7200rpm laptop hard drive>>>>>

Gainging two extra bits is good thing... That's not much more data rate then 8bit. So it would be better to go to 10.

There are small mother boards that can be built into a camera body and run on batteries. Then I could instal either an IDE or sata drive if the board supports sata. The camera inclosure would be in the size of a Panavison 35 Panaflex range. So there is room for components. Ofcourse this is putting the carriage before the horse.

Will the Altsens 2560 or 3560 support 10bit 720p?

[Marto Lautz]

To Steve Nordhauser
Hi steve I would like to know price of the Altasens over gigabit ethernet and if you have a option of start with IBIS5 and then switch to altasens
you can email me at martinlautz@yahoo.com
I'm looking to work at 1080 12bit 24fs 75mbs so I can go stright to pos.
I think you also where talking about multiple interface taht would be very nice.
I think wath you where saying was that the soket/interface is going to be the same to both cmos IBIS5a and Altasens so you yust have to swap sensors.
also how woud you decrive silicon image software?
thanks very much
Martin Lautz

Director of Photography
Filmverlag Productions
www.filmverlag.com

[Zac Stein]

hey all,

I have been sifting through the pages, and a lot of this is going right over my head.

Would it be possible for someone to setup a webpage outlining all the different config's people are using, and maybe the pricing it cost to put one together ourselves.

I just sold my camera and really want to take a step up into something new.

So these seem like a great way to increase my aquisition and picture quality.

It would really be great if someone could outline it all, because atm i am totally lost as different people are all talking about different stuff and i don't know who's config is who's, which clips i am actually watching and so on.

Thank you,

Zac

[Jason Rodriguez]

Hey Markus,

Looking at your latest uploads and I'm very impressed. Looks like something out of Lord of the Rings :)

If you're making a feature out of this stuff, I'm sure it's going to look really good (although I like the non-clipped highlights stuff the best)

About the Altasens,

You should easily be able to have a silent hard-drive array that can acheive over 100MB/s sustained at RAID0. I'm planning on using the new Seagate Momentus 7200.1 series of drives with the SATA configuration. They're not planned for release till 1Q05, but they're going to be very fast. If that doesn't work, then IDE drives at 7200RPM from Hitachi (7K60) will work great too. On the fastest part of the disk they have read/writes of 38MB/s, and they small, silent 2.5" drives.

Another option is the Seagate Savvio 2.5" 10K SCSI drive. Three of these should do the trick as they have read/writes of 52MB/s on the fastest part, and dip down to 40MB/s on the slowest part.

So again, I don't see recording RAW 1080/24p or 1080/30p @ 12-bit as a big problem. If you want faster special-effects frame-rates (slow-mo), then go down to 1280x720 for 60p recording at 12-bits.

With the Varicam, the A/D converter is running at 10-bits, and on the Cinealta it's running at 12-bits. So what's coming off the chip is at 10 or 12 bits, and then the color-controls in the camera are taking that image and processing it down to 8-bits. With 8-bits your dynamic range is severely limited without causing banding. Again you mentioned 8-bits in your digital still camera, but that camera has a 12-bit A/D converter. I've played with the RAW data off the D60 before (not the non-linear RAW's from the file converter, but the black and white Bayer footage, use Dcraw if you want that stuff), and there's A LOT of room to mess around with. That's what the camera has to work with and then it's finally creating a gamma-corrected. compressed JPEG file for output that you're saying creates no problems. But what you're getting of these cameras is a true linear RAW file, and you'll need more than 8bits of information to really work this footage to get those "professional results" that you're talking about. You either have to have some on-camera controls pre bit-depth conversion (like Markus and Sumix do), or you're going to be recording in the top 5 or 6-bits of the image, which will translate to clipped highlights and not a lot of room to fix them (and if you try to color-balance you'll get some sweet banding artifacts, especially in the shadows ;-)

Another note on the JPEG'd Viper material. That's a LOG encoded file, giving you at 8-bits the same visually perceived bit depth as a 10-11 bit linear image (at 10-bit LOG you have the same visually perceived bit depth as a 14-bit linear image). So to say that you can "work" with that nicely is an understatement. An 8-bit linear file will NOT treat you as nicely or be as much of a joy to manipulate.

[Michael Pappas]

Jason Rodriguez, how would this spec be then 720p @ 24fps in 10bit?

At this level it's 26.3 MB/sec this still would allow the use of a portable PC or a built in PC mini mother board into a camera rig.

Michael Pappas

[Wayne Morellini]

A lot of these questions have previously been covered in the threads, and as an old hand here I would like to answer them for everybody.

Zac:

There is a web site that Rob.S has setup with an updatable wiki to keep track of components and configurations, so you can ask Rob and he can give you access to update that:

http://www.obscuracam.com/

If anybody wants to go through all the threads and update the wiki with all the factual information (products, systems, workflow and software solutions) and separately theoretical, your welcome, just email Rob.

My Technical discussion thread has links to all the threads in the first post.

Markus:

What you said about large storage requirements of 12 bit 1080 is true, 8 bit just requires a person on the camera to set the cameras up for the scene, like any normal movie camera, and as I understand your camera is already reducing 10bit log to 8 bits. Personally I would prefer to use 10bits+ myself for Film and do a lot in post, and ride the settings live for on the spot television (I am one of the few interested in high quality documentary). Theoretically the scheme I would like to use instead is record straight footage (maybe adjusting to maximise range) but riding the image settings in the view finder and then recording those settings with the straight footage as a suggested reference. This then can be used as an quick automatic reference for post image processing, where the expert operator (in a nice calm environment) can interrupt and further adjust it where it needs to be enhanced.

Viper is an old system now and the economics and what can be done has moved on from when it was introduced. There has been solutions suggested for the large amount of storage, one alternative was to back it up on server backup tape, expensive but probably cheaper than X amount of damageable hard drives hanging around. The fastest drives will do over 70MB/s sustained (can somebody verify that??), two drives are not going to kill the bank, a mini motherboard, battery and a sound proof case still adds up to something that can fit into a PD150, smaller than a large movie camera. But with in camera lossless compression your data rate drops in half, allowing 1080p on cheaper drives. The interesting thing now is that visually lossless codecs (which I would use in everything that doesn't require special effects, or some IMAX like resolution upscaling) will do 4:1 on bayer (though they did do 6-10:1 on non bayer)(Avid also has a low compression rate codec that it used on the new Ikegema camera). If we get computers or FPGA that can process that than storage can be further reduced in future.

With the other portability and size problem, you can use Giga Ethernet as suggested, but the new Ultra Wide Band can run standard USB2.0, and Firewire A, wirelessly over 30 feet, so there is another possibility (in future, it would probably need to be adapted to suit camera data, or cameras built that suit it, because of more errors in wireless data) the option to have a small head wirelessly linked to a remote big computer.

But you have done the right thing, and used the most balanced 720p solution that can be currently put together (maybe there will be a super version B of the IBIS one day (though the A version is significantly better than the first one).

Michael:

Your configuration for 720p camera is good, and what Ben has said is good, but here is the rub. Whatever hard drive your laptop has you have to check for sustained data rate, which is the fastest rate it can guarantee to do all the time without overheating the drive head or unit. There are many things that cause a drive to slow down from it's maximum data rate even going from the inside disk track to the outside will have to different speeds) (the OS and system being very big culprits too, that is why capture software has to be programmed around them for lower speed systems). Big ram and fast drives is the way to go (the ram is used to buffer data and the system software, to allow the capture to have the maximum efficiency). Now sustained is going to be a lot less than max, and the max is usually a lot less than the interface (all that 133MB/S interface stuff doesn't help very much when the max and sustained are less than 50MB/s). So carefully check, I assume that the numbers Ben quoted are the sustained values, older drives are a lot slower. There are storage sites that reviews these figures on drives (but I don't know about laptop drivers) that have been mentioned in the threads before (storage.org is one, I think).

Now with how much data 10 or 12 bits has compared to 8 bits, and I'll throw in the shutter for everybody. If you are using a memory buffered capture device then shutter doesn't really effect capture, but if your not then twice as fast shutter doubles the amount of data that has to go across the interface or capture card. Also the frames are read out slightly faster than the shutter. Now if the camera doesn't pack data properly (particularly cameralink) any pixel over 8 bits in size will occupy 16 bits instead. On a, cheap, unbuffered interface of PCI you get max 80-100MB/s Sustained , USB2.0 and Firewire A around 50MB/s Sustained. Now on an unbuffered/unpacked camera, an approx 24MB/s (+ slightly faster frame readout) bayer 720p, 24FPS, 48th/s, shutter 8 bit needs approx 48MB/s+, you just about maxed out the USB2.0 (also processor hog) and Firewire A interfaces. If you use 10 bits, you have approx 96MB/s and maxed out PCI. So there is no question of using faster shutters. If you get a camera that buffers, packs and compresses there will be some cpu loss to unpacking and decompressing for preview (but if you have the right preview software much better decompression hardware are on machines than compression hardware). So many people are going for 64-bit PCI Cameralink capture cards, that have at least double the data rate of normal PCI/Giga Ethernet, but on a laptop you are limited to 100MB/s Gigae (with specific Intel Gigae chip and driver) or 80MB/s expensive cameralink cardbuss framegrabber (I think they are as much as a laptop, anybody??) . But as far as disk space and data rate goes, you can pack and compress before you send it there.

Proven cameras, well apart from what is used mainly by the broadcast camera companies, Altasens. The IBIS is nice and cheap but I would imagine not as good as the expensive, compressed, HD broadcast ENG. The SI micron camera that Obin was using has better pickup/colour, but smear etc and global shutter :( (new version please). Actually a question for Steve, are they going to release a new version or 720p capable camera? The 3MP 1080p Micron camera from SI is problem free, but requires a lot more light. Now there are a lot of cameras out there, they usually are Micron, IBIS, or CCD chips from the major broadcast camera manufacturers, or now Altasens, so choice is limited in cmos.

If your looking for everybody else's clips they have previously been posted, mostly here. You could probably search for them (select search by page option, feed in thread title and do www OR ftp OR http OR whatever extensions). I think Obin and most of the usuals must be busy at the moment. I currently am working on/finishing some political stuff, and am just feeding in some stuff before I pack up and go on a long break from all this, so I won;'t be around for to much either.

Anyway with compression/packing/buffering I think that you can go 12bit 4:4:4, 1080.

We have discussed mini-PC boards in time past (P4, PM, VIA, Mini ITX (17cm*17cm) and nano ITX (12cm*12cm/8cm??), all the Intel small formfactors, and FPGA (there was a thread for that even).

If anybody would like to do something, please do, and discuss it with us. If anybody wants to summarise on the Wiki, Laurence you might like that, so that everybody else can read it instead of the threads, please do. Otherwise read the threads and discuss. The problem at the moment is the threads are too big to read and the same issues get asked so many times that I guess it is too tiring for most to answer them again, if only there was a faq post/or link available at the start of every technical/Alternative imaging thread over 80 posts ;) ?

Rob, I notice that you have been really busy on the Developement Blog, how's things going?

Thanks

Wayne.

[Flax Johnson]

Hi,

I'm following this thread with a lot of attention and clearly I would like to develop my own HD camera solution.

As a software engineer, i'm not afraid by doing bayer filtering or handle this amount of raw data. (May be I'm wrong !)

My main question is about which sensors/implementations to use (I'm a little bit lost) ?

If I believe what Markus (nice job man) said, the global shutter seems to be not the best solution, indeed.

Can we go now with the existing sensors/solutions provided by silicon imaging for example or is it better to wait for some months.

ps : I could be interesting to join a starting project. My goal is to shoot a movie much more than doing homeworks after a full day of programming for a living.

[Jason Rodriguez]

BTW, if you adobt the DNG format from Adobe, you won't have to worry about bayer filtering (let Photoshop, Phase One, etc. do it for you in a very nice interface)!

Also Wayne, while the 16-bits for anything over 8-bits is true with the EPIX cards, with the GigE adapter, there's a buffer on the adapter that now supports packed bits, so you can transfer only 12-bits from the adapter over the gigabit ethernet line, which can give you up to 30fps @ 1080 with 12-bits (800Mbs transfer rate over GigE).

Also for the preview software, instead of decompressing the 12-bits, either use a look-up table to map it to 8-bit, or just clip off the top 4 bits and look at the bottom 8 bits (or say the middle 8-bits, chop off the top two and bottom two bits). The "clipped-bit" setting shouldn't take much processor resources at all.

Another thing to beware of is the fact that the Pentium M is a very fast little processor. It's SPECINT2000 scores are 1500+!! SPECFP2000 was a little over 1000, again, nothing to sneeze at for such a small little package and plenty of horsepower to do what we need.

[Flax Johnson]

Thanks for the infos Jason.

Will you go for IBIS or SI solutions with for example the upcoming 1920.

Btw, I was looking for a dual proc. mother board.

[Aaron Shaw]

BTW, if you adobt the DNG format from Adobe, you won't have to worry about bayer filtering (let Photoshop, Phase One, etc. do it for you in a very nice interface)!

Really?! How would that work? I've never used the DNG format before.

All of this really is amazing. I can't wait until we get a fully functional Altasens cinema camera working with the bugs worked out. I'd buy one in a heartbeat.

Where do you guys think we will be two years from now? Will you guys still be working on improving your cinema cameras or do you think some of the larger companies will actually give in and give the public what it wants? I'm probably not going to need a camera for another year or so and I am very excited about the prospects!

[Rob Lohman]

The images look great Rai and Markus, thanks for sharing them.

So for now we know that:

Chip: IBIS5A with some unknown (custom?) D/A converter
Resolution: 1280 x 720 x 8 bits @ 24 fps
Media: harddisk

Are you guys going to put up any pictures of the camera? Would
be interesting to take a look at how it is together with a RAW
frame grab either in uncompressed BMP or before bayering.

Michael Pappas: it's great to have you onboard!

Flax: global shutter IS the BEST way to go. The other option is
rolling shutter and that gives you a skew on the image if either
the camera is moving (fast) or your subject(s) is/are.

Steve: did you say in an earlier post that you guys have
implemented your own A/D converter on your IBIS5A product
to insure a better SNR?

Aaron: DNG is like RAW file format from Canon (basically). So yes,
photoshop allows you to do a debayer. The only issue is that it
will (I assume) only do this one frame at a time. That's not good
for movie work...

Jason: you can't easily clip 10 or 12 bits to 8 bits. I tried that and
got strange results. Here's why, consider the following 10 bit value:

11 0000 1101

If you strip the top two bits you get 1101, which is a much, MUCH
darker pixel than the original color, so you would at least need
to make sure the high bit is set on the 8 bits if the 9th of 10th bit
was set.

[Steve Nordhauser]

Rob:
We use a 12 bit A/D so statistically you get more dynamic range, if you clean up the image (offset and gain correction). You also get better response than the sensor A/D at higher clock rates SNR at 24fps (depending on the actual clock rate) might be the same.

The issue with the global shutter on the IBIS-5A (OK there are several) is that it can't overlap integration and readout. This means if you run the clock at 22MHz (1280x720x24fps), you would get no exposure time (about right for flash applications). If you want 1/48th sec exposure times, you can expose for 1/48th and readout in 1/48th (45MHz clock) to get a 24fps frame rate but that is really pushing the internal A/D - that is why we do the external. Of course 30fps is only worse. The first thing you will see is color impurity - pixel level smearing that gives the sensor a washed out, low contrast look.

The IBIS-5A also runs in rolling shutter mode and can overlap the integration and readout, but then all you are gaining over the Micron is the larger format (2/3 instead of 1/2) in trade for noise and sensitivity with the Micron cameras costing less.

[Jason Rodriguez]

Idealy though I would want a LUT to map from 12-bits to 8-bit for the screen, but I'm not sure how long that would take to map out.

What if you just took the top 8-bits for preview, is that computationally easy? What are other software that are packaged with frame-grabbers doing? I remember at NAB going over to the Altasens booth, and they were showing an HD image on the screen playing back in real-time from the camera (in this case a 3560 demo unit) in 1920x1080. So playback is possible, but I'm not sure if they're throwing away bits or what to get the real-time fast playback.

Also I guess playback only needs to update at around 24-30fps, anything else I think is overkill for playback preview, and it has a comutational cost on top of "typical" playback speeds.

[Jason Rodriguez]

The only issue is that it will (I assume) only do this one frame at a time. That's not good
for movie work...

Photoshop can do batch conversions (you have to set up an action), as well as Phase One's DSLR Pro (which has very nice interface for tweaking colors, white balance, sharpening, etc.). The nice thing about DSLR Pro is that it can also do background conversions while you operate on new scenes in the foreground, so it maximizes the efficiency of your machine while still giving you great images.