4:4:4 10bit single CMOS HD project

[Jason Rodriguez]

Hey Obin,

From what I've seen of the Altasens, it's far beyond the 3300 in ability. Saturation is not the only indication of quality, you can saturate something as much as you want if you have the right software. What does make a difference is the amount of noise and bit-dept; that's what limits saturation and other color-correction processes.

Frankly, from my experience the Altasens equals or surpasses the Thomson Viper in noise levels and dynamic range.

Of course it would be nice if I had a better RAW/bayer converter, such as something using Variable Number of Gradients, or maybe the one David Newman was working on, etc. Right now the converter I'm using, while quite functional, has a lot of problems when there's the slightest bit of noise, producing gridding patterns, etc. and I usually have to add a gaussian blur filter to remove that artifacting, and then re-sharpen the image afterwards.

But for real, the amount you can push this chips is quite astounding. And it pretty much will achive it's theoretical limit of 10 f-stops (1000:1) with the amount of noise that I'm used to seeing in a digital still camera like the D60 at around ISO500 (that would be ISO400 plus another 1/3rd stop). Or again, the amount of noise in the Viper, F900, etc. when trying to go for the maximum dynamic range (they're all around the same amount of noise, and to get the maximum dynamic range you have to underexpose them all a little, so again, it's around the same amount of noise or a bit less). Very good noise levels, color, and good dynamic range. I don't think you're going to get that out of the 10-bits on the 3300. In order to do that, you're going to have to set the white chip on the Macbeth to a setting of 25 RGB (out of 255 being maximum white), and at 10-bits, you simply can't do that without some severe banding when you normalize that, taking the bottom 10% of the image and stretching it out to the top 80%-90% and compressing the rest into overwhites. The clean 12-bits on the Altasens can do that, and I've been very impressed with this capability.

But whatever the case, as of right now, to get the quality we want out of these chips, we need a top-quality, big-screen capable bayer-converter; or maybe somebody has one and isn't telling us ;-)

[Jason Rodriguez]

Obin,

You got any screen shots of your capture app yet? Also is yours just a capture app, or a convert app also, and if so, what bayer de-mosaic algorithm are you using?

[Rob LaPoint]

Hey Steve, should we be worried about initial quantities of the Altasens camera when you guys actually release it? Have you been excepting orders or do you not anticipate a problem? I can't wait to actually get my hands dirty on this project but unfortunatly I don't have the money to buy a 3300 just for experimentation.

[Steve Nordhauser]

Altasens:
I don't see any real problems with the Altasens availability....once they start shipping. The problem is the turnaround time for silicon. They solve their problems, simulate, change masks, queue up for the foundary (no one actually makes their own chips anymore) cut and test the dies, package and test. This is an 8-16 week cycle. Once they have parts packaged, there shouldn't be an issue (I hope, since we have a bunch of orders on the books).

Obin:
I'm not sure at the raw data level what better saturation means. Better response curves on the Bayer color filters? The two sensors are from different companies so there is a good chance their filtering is different. Better IR cut filters? I believe Altasens suggested some and we would go with their suggestions. Slow pixel response (slow analog path) smearing one color into another? I think that both are OK with this but it is easy to tell with pure R, G, B targets. To be fair, you are comparing cameras that were meant for different markets (even at the sensor level) and a 2X price difference. If Altasens knows what they are doing, and they seem to, it will be a much better camera for those who can afford it.

Ethernet:
I just finished some more instruction on the ethernet interface and was given a benchmark on the SI-3300RGB. It was running 1920x1080@24fps, 10 bits (packed as 12) over a gigabit interface, continuous with 24fps *color* data display. Pitiful algorithm but you would record in raw mode. This is about 600Mbps. This is not cheaper than the camera link solutions but I would toss it out there for you integrators.

Jason:
Thanks for the work on the SI-1920 - no one flip out - he doesn't have one - but has been testing some sample images for noise and dynamic range for us.

[Jason Rodriguez]

Hey Steve,

Thanks for all the info. Also a lot of thanks to Ari Dresler who was willing to put up with my frequent requests for images at many different setting, etc. when I meant for him to poke around with registers on a "not-so-easy-to-use" interface.

And yah, it does seem like Altasens knows what they're doing, there aren't many motion picture sensors (practically none actually, except for the Viper/Cinealta, probably the new Panvision and prototype Arriflex) out there that can take a signal where the white chip is recorded at 25 linear (out of 255, or 10 IRE on a scale of 0 to 100 before gamma correction), and then get a wonderful, practically noise-free image (again probably equivalent to a pro-digital camera like the D60, 10D, D70, etc. at ISO500) with good color saturation. 9-10 stop range no problem. Quite amazing, and very exciting for us.

[Obin Olson]

So we have passed the worst now...all this time our mother board has been the big hold-up! the pci bus has a problem and has been causing us weeks of run around and crap..I got a new board and I am running the 3300RGB at full 12bit 1080x1920 30fps!! not on disk yet but soon!

man it takes a pissload of light for this 3300 camera! I am not sure we could even use a mini35 type adaptor with this chip...how will the Altasense be?

[Jason Rodriguez]

Obin, you shouldn't be trying to expose the white chip (on a Macbeth chart) for 90% white, that's too bright. If you expose the white chip for 50% white, or maybe 40%, you should get more "sensitivity".

[Obin Olson]

what is a "white chip" Jason?

[Jason Rodriguez]

The 90% white chip on the Macbeth. At the bottom of the Macbeth chart there are six grey-scale chips. The 90% white chip is of course the "white" looking one on the left-hand side.

Of course if you expose the 90% chip at 40-50%, you're technically underexposing, but that's what the internal camera electonics are typically doing anyways to create the headroom needed for soft overexposure clipping and extreme highlight handling in professional video cameras (i.e., those "film-gamma" settings that are advertised with certain HD cameras like the Varicam). I'm figuring that the 3300 shouldn't have too much noise so you can safely underexpose at those levels and then normalize/color-correct in post.

[Juan M. M. Fiebelkorn]

Jason please call it White Patch to avoid confusions. ;)

Also if someone is interested about discovering what is the exact exposure for a given light situation and have a Macbeth chart , here you got the corresponding RGB 8 bit values for every patch on the chart:

http://www.brucelindbloom.com/index....heckerRGB.html

So, if your exposure is right you should get the more or less same values after conversion to 8 bit with a gamma correction of 2.2 on a PC..

For informative purposes I will tell you that value for Black Patch is around 50 and for White around 241....
This doesn't mean that your blackest black will be 50, remember the Chart is a piece of paper and it reflects some light.If you put a piece of black velvet you will get a deeper black..

[Obin Olson]

can someone give me a link to a chart that I could print out and shoot? something that will show dynamic range from the 3300rgb?


http://www.dv3productions.com/test_i...outside CC.jpg
http://www.dv3productions.com/test_i.../flower CC.jpg
http://www.dv3productions.com/test_i...-flowers-2.tif


all from 3300RGB Silicon Imaging Camera..I am happy with the images..I think it takes very good images and with the plus of having no smear and 1080P this is a good unit..I hope to Have CInelink Working with it soon at 24fps!

all captures above are with Xcap and the very BAD bayer filter it has ;)

it does take lots of light ...this chip/camera will need to be "lit" like film...this is a good thing right? keep the production value high! ;)

FYI the Epix capture card does NOT work with the Asus P4P 800 boards! even though the board is running the INtel 865PE chipset ..I don't know why but I have wasted weeks with this crap and I would hate anyone else to do the same!

Now I have a Gigabyte board with SIS chipset and it runs fine ..never FIFO overflow even at the max 1080p 60mhz clock!

Steve what is the MAX clock this 3300rgb can run at? 85mhz? 75mhz? I need to get CIneLInk running at that max clock all the time to reduce rolling shutter ;)

[Jason Rodriguez]

Hey Obin,

Why do you need to run at the Maximum Hz, when all you need is 48fps and dropping every other frame to get the same motion characteristics as a film camera with a 180 degree shutter? If you run the chip too high, you're not going to get the longer 1/48th of a second shutter you need, and your images will look really choppy, like "Saving Private Ryan" or something like that. Realistically, there shouldn't be any rolling sutter artifacts when you're running the chip at whatever Hz is needed for 48fps. I believe for the Altasens that we'll need to run the chip a 58Hz for that 48fps threshold.

[Obin Olson]

because you can run at whatever fps you want with a high mhz..that is how we do it.

put the mhz at 60-80 for ALL shooting and do what you need with the shutter speed and or fps

[Juan M. M. Fiebelkorn]

Obin,
Could you give me the uncorrected 16 bit version of the outside image?
I mean the color image linear (not gamma corrected)....

[Obin Olson]

http://www.dv3productions.com/test_i...16bit 1080.tif

this one?