4:4:4 10bit single CMOS HD project

[Rob Scott]

Juan M. M. Fiebelkorn wrote:
wxWindows ... runs on almost every available OS

I've heard good things about it, and I'm not sure why I hadn't considered it for this project. I'll look into that one too. Thanks!
(Easy to use = A Good Thing)

[Juan M. M. Fiebelkorn]

Could someone tell me anything about the SI-3170-CL ????
It is supposed to give 2048x1536 @ 30 fps.

Why nobody talks about it?

[Jason Rodriguez]

Sounds good,

BTW, do you know what type of bayer algorithm you're going to incorporate? Variable Gradients, spline-based, Optimal recovery, etc.?

One thing I might mention is the inclusion of some sort of "blur" or filtering effect that can be variably applied (variable from none to a decent amount depending on the image, not really a gaussian blur, but some sort of anti-aliasing type blurring filter so it doesn't just create overall "softness" like a gaussian blur does, but it does soften to reduce edge artifacting or other bayer-artifacts). IMHO I think Bayer processing can really be a weak link in this whole system because that's where you get your final image from. If you don't have a good bayer algorithm, then you can have all the bit-depth you want, but you're going to be fighting all those nasty bayer artifacts that can be quite distracting, and just scream out the fact that you shot on digital.

BTW, one more thing,

Obin, the motherboard we were looking at from Kontron (the P M model with 64-bit PCI-X), how would you hook a battery up to it if it requires an ATX-type powersupply? Can that even be done without making the mechanism a behemoth? Just curious how that might be done.

[Wayne Morellini]

Little PC nice, case, could be made to look like a shoulder cam. If you have a look through this thread, the viper thread ands my Custom camera thread, you will find references to simular devices.

There is some miss conceptions here.

For 720p RAW, 1Ghz should be enough, using MMX, SSE, GPU, the new media processing architechure that via seems to be developing, and whatever else we can milk, maybe even simple lossless routines (Huffyuv ??). PentM claims simular power consumption at 1 Ghz, so maybe it is as good, or better.

Just because it is on an itx board, doesn't mean it can't use a laptop chipset.

As I said before, whatever board and whatever case you want, the future software is meant to support a wide varitey of configurations. If anybody wants to use a laptop great, do it. But there are other people we also want to consider and serve aswell, and we are trying to do cheap entry level and professional versions for them.

Looking at the big picture, thinking of the majority of people out there, and growing this market:

• We want people, for the project to be a sucess, rather than a red herring, that will also give involvement, supporting products, and volume.
  
  To get success we need it to appeal to the most people. For this we need it to be simple, reliable, professional, cheap and with good quality performance.
  
  The majority of people are not techy computer people. Techy people, and their circles of techy/very intelleigent freinds, often miss this piont and try to get everybody else to conform to how good they are.
  
  A big portion of people are video/film people rather than computer people, and don't want the computer hassels or the techy image, they want simple, reliable, plug and play.
  
  People who are video/film people might mock at the techy idea of carrying a laptop around instead of a easier professional END shoulder case, and not support it.
  
  Raise the cost and you loose most of the wanabees, prosumers, and low end local professionals (wedding videos etc). So we want cheap non laptop versions for these people aswell.

Earlier I was writing:

It might be splitting expensive hairs to use a laptop instead. The via embedded, and Antaur laptop, chips/chipset are very close to laptop versions. Unless you use the laptop intact, it is not going to be simpler. If you do, then how easy is it going to be to carry the laptop around (on shoulder strap, waist, backpack etc) corded to a camera head, then using an ENG shoulder case. To warp the laptop to another shape case you will have to break it apart and use some longer cables. It will be more expensive, and the laptop drive might be a low data rate drive that needs to be replaced with a desktop drive. And the laptop main baord may not even be much smaller than the nano itx (12*12cm size, but I think I heard of 12*8cm aswell). The other problem is that unless your using USB2, a pccard cameralink will have 20MB's less bandwidth than mini-pci. The truth is that a good nano itx motherboard will be just as low profile and low powered (or not a significant enough difference) as laptop boards except they are cheaper. If you use the PCI-E desktop bus derivative of PCCard (formerly Newcard) then you will have 266MB/s, enough for basic 24fps/24th sec shutter (48th sec requires twice the bandwith on most sensors) 8mp 8bit shooting, or 100fps/100th sec slow motion 1080 shots (10bit with require twice as much bandwidth unless packing is used) which could also be used as an electronic exposure.

I am focusing on a few cases. A small sized allweather/underwater shoulder case (I want to give it long sexy lines simular to, but not the same, as the Olympus SHD case), nano itx (12cm*12cm) is preferable. A whacky styled (move over XL1) Indie special for upto 17cm*17cm boards (or newer simular sized Intel), the case for my camera project last year. The last one is a handheld model (upto PD150 in size, but handycam would be preferable) nano itx sized (or handheld portable PC). The allweather case can be designed so that any small pc, like the Little PC mentioned, can be slid inside, but this will defeat the unique cooling system, so it might need to be dropped. Some of the cases could be made to accomodate upto micro-atx size, but any bigger you might as well goto fullsized cases and dolly for them. But the work on these has yet to be started.

Originally posted by Jason Rodriguez : A 2Ghz Pentium M is very fast when you turn off all the lower-power settings. A 1.6Ghz Pentium M was out-doing many 2.4Ghz P4 Desktop systems on AE benchmarks, Cinebench, etc., so at 2Ghz, I'd expect even better performance.

Very impressive, it is amazing what outside of Intel designs can achieve (they bought the Israeli company that developed the Pent M). If 1 Ghz VIA is 7w, smae as a 1Ghz Pent M, then we should also closely watch the Pent M as well. But the question is how does the MMX, SSE etc modes perform on the Pent M compared to the P$(4) that is supposed to have a special enhanced architecher for media processing. Does Intel get beat again?

Originally posted by Ben Syverson :Wayne, as we were discussing on other threads, I don't think the VIA systems can handle raw 720 @ 24fps. At least not yet. Maybe they'll come out with a kickin' 2ghz system at some point, and we can take another look. But for now, their 1ghz fanless nano boards aren't even out yet and they will be underpowered for our purposes.

The software isn't ready yet, we are waiting a few months, and I think the 1Ghz nano-itx were announced a week or so ago. Use a laptop now, it doesn't matter.

On a futuristic, pie in the sky, aliens abducted my XL1 type of side note. Interesting article in New Scienctist last week.

A researcher was claiming they could, one day, make a device that uses Ultrawide band technology to deliver between 100 to 1000 Gb/s that could be held in the palm oif your hand (and probably be used as a tricorder I would say). Ultra HD could barely dent that.

[Juan M. M. Fiebelkorn]

Well, I guess that the SI-3170-CL is the camera Steve N. was talking about (but he called it SI-3300-CL).
Anyway here is the datasheet from a German site that is offering it now!!

http://www.ehd.de/products/cmoscamer...MegaCamera.pdf

What I still can't get is: if it is the Micron sensor or not.

[Rob Scott]

Juan M. M. Fiebelkorn wrote:
SI-3170-CL / SI-3300-CL ... What I still can't get is: if it is the Micron sensor or not.

Check out the SiliconImaging site. The SI-3170-CL and the SI-3300-CL are different cameras.

SI-3170-CL:

• 2048 x 1536 Resolution
• 1/2” Imaging Format, 3.3um Pixel
• 12 Bits per Pixel (Dual-Tap)
• 100Mpixels/sec Throughput
• 30 FPS Sequence Capture at full resolution

SI-3300-CL:

• 2048 x 1536 Resolution
• 1/2” Imaging Format , 3.2um Square Pixel
• Rolling Shutter, Progressive scan
• 10 Bits per Pixel, 48MHz Sampling
• QXGA(12fps), UXGA(20fps), SXGA(27fps) & XGA(43fps)

Note that the 3300 can get only 12 fps at its full resolution.

[Juan M. M. Fiebelkorn]

So?

There comes my confusion.....

Steve wrote a couple of times 3300, then appears the SI-1920HD and so on..

And I still can't understand why we aren't talking about the SI-3170CL

[Wayne Morellini]

I accidentaly posted to the wrong thread before:

Originally posted by Obin Olson : Juan, is that 8bit only? we really need 10bit for capture/compression - then we can bump down for 8bit editing later on...if we need anyway...

It should be a lot easier to extend this to 10+ bits in code or FPGA, then design a new one.

I agree with Jaun, with MMX, SSE etc and all maybe Huffy canbe made to work (impresive it gets so far without it).

It is best to keep it in bayer for master accuracy aswell.

Thanks

Wayne.

[Rob Scott]

Juan M. M. Fiebelkorn wrote:
And I still can't understand why we aren't talking about the SI-3170CL

Perhaps Steve accidentally said 3300, because it really is not suitable at all.

We have talked about the 3170 a lot, and Obin was considering it. The problem is that scaling from 2048 -> 1920 is not a simple pixel "binning" proces, it requires interpolation. Or else you have to "window" it -- losing some pixels, but worse, you have a smaller chip area so your DOF is worse as well.

For HD, the AltaSens (SI-1920) is definitely a better choice.

[Juan M. M. Fiebelkorn]

Oh, thank you Rob, indeed.I was really lost :).

I've been testing the Elphel's camera 313 wich uses the same Micron sensors, and it is very sensitive, really.
If you set an exposure higher than 10 ms, everything gets blown out!!!!.

What would be the problem of using it at 2048x1536?
Anyway we need to apply heavy processing on the images after capture to De-Bayer them, so adding to this a simple crop and a Lanczos resize to 1920 won't be bad.
The other advantage is we could use an anamorphic adapter anytime!!!!! ;)


hahahahaha, sorry.I see the problem now!!
It would give as at least 108 MB per second (suppossing we get only 12 bits and not 16) @24 fps.

[Rob Scott]

Juan M. M. Fiebelkorn wrote:
It would give as at least 108 MB per second

Yes, the data rate would certainly be higher. The image quality might be marginally better since you'd be downscaling from a higher resolution after the Bayer filter. But I suspect that the AltaSens chip is going to be better quality overall anyway.

[David Newman]

<<<-- Originally posted by Juan M. M. Fiebelkorn : A question to anybody who knows about the topic:

What could happen if I store Luma info with 10-16 bit depth and U and V with 8 bit depth?

What could be the final bit depht when converted back to RGB?? -->>>

This can be a good approach -- I'm considering it myself. Increasing luma bit depth without chroma increase in precision will result in cleaner gradients and better dynamic range in shadows and hightlight where chroma subtly is less import. 8 bit chroma will limit you a small amount for color corrections that involve increasing the saturation.

[Juan M. M. Fiebelkorn]

David,
But when we go from YUV to RGB, U and V planes are merged again with some portion of the higher bit depth Luma, to get the Red and Blue channels again.
So from this very moment, Red and Blue can't be 8 bit depth anymore......My thought..

[David Newman]

There is no direct effective bit depth due to this conversion, but you are correct that red and blue will gain additional precision is luma precision is improved. Increasing luma alone does provide more quality increase than increasing chroma alone.

[Jason Rodriguez]

Do we have to go with compression? Can't we compress after the fact so that those who want uncompressed high-bit depth can have it?

If I wanted compression I'd shoot tape! :-)