4:4:4 10bit single CMOS HD project

Obin Olson · July 11th, 2004, 10:40 AM

I like the look of the Zeiss lens I think I may bid on that one!

Obin Olson · July 11th, 2004, 10:42 AM

Steve how can I tell the junk on ebay is the right size - next the ebayer seller will not take it back ;)

Wayne Morellini · July 11th, 2004, 10:53 AM

Quote:

Originally posted by Juan M. M. Fiebelkorn : I'm feeling sad at this moment cause I see the technical part of this thread seems to go nowhere most of the time....
Anyway here I go again.

I'm trying to develop an opensource project for a codec for this project.The main idea is to use a mini-itx motherboard with a P4 or athlon (may be a mobile athlon).Now the codec is performing more or less good, giving

There are also some huffman compression working FPGA projects, which I posted before but nobody seems to notice them.

Yes this quiet often happens with the technical discussion. See my recent post about the interveiw at linux devices with the VIA processor cheif, he is releasing the next version with extra processing for aps that sound like comrpession, so you can skip the power hungry P4, FPGA etc. Nvidia, with the Nforce4 chipset, may have 20 Gflop dsp power ready for use with comrpression (aswell as the 3d pixel processor), and other companies, this is most of the compression power you would need, other companies might do simular, and maybe Shuttle, or somebody, will do a small, version of it. The DSP in it is their sound chip, and I think I heard previouse versions were availble for other DSP purposes.

I have February issue of "Elektor Electronics" and they have a series article "Hands-ON CPLDs" and they have information and links, but I thought I had "Electronics and Wireless World"?? UK that has a feild programmabe tutorial series as well at the moment with links to teaching aids and advice.

www.xilinx.com
www.coolpld.com
www.altera.com
www.latticesemi.com
www.elektor-electronics.co.uk/dl/dl.htm
(no more links in that months article, I think I have seen better ones).

I've been thinking about codec tonight (I was planning one for my OS system) and jelled a few more ideas together. Using ideas from the OS's planned 3D compression, and audio/vioce compression systems, I think much higher comrpession should be available. Once you eliminate the random niose and restore the approximate origional I don't see why you couldn't average 10:1 lossless compression (maybe more). I don't even think about these draft ideas much because they are supposed to come out in generation two or three of my OS, and until I have the money for the intellectural property issues (patents and licensing etc) it is just a mindfeild to do it, and I need to keep control of it so I am free to use it on the OS. This sort of stuff could take years to do, my plans go way past the norm and require much more horepower than MPeg 4 compression.

What do you guys think I should do, the clocks ticking and I'm not getting any younger, richer or much weller?

Quote:

Originally posted by Steve Nordhauser : OK, I was hinting that I might have something interesting brewing. Micron has a 3.2Mpix sensor that is part of the same family as the 1.3Mpix that Obin and Scott are using. The

This is good, is this the micron 40 chip they were talking about before.

Obviouse Questions: Is it really much worse than the altsens chip, in what way, and can we see the specs? I think I might prefer the extra detail to a bayer Altsens 1080 design. You mentioned Gigabit Ethernet, is it going to be on this camera, that should take 34fps, full frame, is there subsampling to get 1080, 720 at 50fps?

I like the propect of 3 chip (though not the HDD space) 1080, or even 720p, or 8Mp bayer. So what is the possiblity of a 3 chip version on this chip? We are talking about using 35mm SLR lense adaptors anyway, so only the low light, SN, and latitude are of concern for me. Somebody mentioned that prisms for three chip could be had for as little as $500 or a $1K, when is Foveon ever going to come to my rescue? Also my question on dual slope mode, how does it work and can the results be alligned to produce a natual continuose contrast image?

Very Thanks

Wayne.

Quote:

Originally posted by Matthew Miller : Steve,

From the back of the room, the images from your camera and the images from all the other clips matched up just fine. From seven feet away, the grain in the stuff shot on actual 35mm film is very apparent, where as your clip is free of any grain. The final shot of your face in the wmv file was the point where everyone commented on the fine detail of the image.
Grain is added digitally all the time, espescially for composited CG work in movies. Still, the slight doubled

Because I'm into designs for all sorts of things, I have done some simple testing on matching the size of the screen. If you have smallish lensed glasses, close one eye and move towards the screen with one side of the screen on the outside until the screen fills 3/4 of the frame horizontally, this is approximately near the start of the confortable veiwing distance of the second third of s small commecial cinema. For medium width lense it is around 2/3 (the ones I developed it with), for widish glasses it is nearer 1/2 of the lense. It is around 20cm from a 17cm monitor (hard to see). Get a cheap pair of sun glasses and try it out. Any closer than this your in the cinema seats most people avoide anyway, so it is a good distance for judging subjective quality.

I like HD because it doesn't have grain, but it hides imperfection anyway.

Obin Olson · July 11th, 2004, 11:00 AM

Wayne why do I find it hard to understand what you say in your posts?

is it just me??

Wayne Morellini · July 11th, 2004, 11:20 AM

Quote:

Originally posted by Les Dit : Yes, with the HD10 people were actually 'seeing things', as it turned out. It's simple to still frame it and look for the double frame artifact. None was found.

In my case I wasn't, the 30fps, 60fpp footage was being down converted display on no 60Hz rates.

Quote:

Originally posted by Steve Nordhauser : SI-3300:
You can't subsample with this other than is even multiples so you can't get to 1280x720 since the basic sensor is 2048x1536. This means you have to use ROI (region of interest - windowing). The small pixels mean

Doesn't matter, a bit more than 1080, Electronic Image stabilisation, Rob?

This definetly isn't the 2/3" Micron MI-MV40 4MPixel that Sumix was talking about before, what happened to that one?

Is it this one, some more information, serial ATA??
From viper thread:

Quote:

Originally posted by Adrian White :
Last week I rang a guy called Steve Nordhausen who works for silicon imaging. Within 8-12 weeks time a new camera will arrive! Here is the spec so far, forgive me if they are not detailed enough.

SI-1920 HD camera. 1920*1080 3.2 megapixel at 24fps poss 23.976 as well. Single cmos chip. Will stream to computer (PC) poss USB2 connection altough serial ata configuration was hinted at.
Frame grabber software will be required. Apparently 10bit and 12bit solutions are available. Images will be uncompressed. Frame grabber will be required along with another piece of software which I'm still a bit hazy on. Ready for the best bit? Camera will be approx $4000!
Frame grabber $1500. Camera will apparently be compatible with 16mm bolex or sneider lenses along with others. Steve mentioned that a press release would be availble close to release which may be July!

I would appreciate hearing what you guys think about this. Personally I don't think it's vapourware as they already have a 6.6megapixel camera that does the same but at different frame rates. Look forward to hearing from you guys.

Obin Olson · July 11th, 2004, 11:48 AM

see what I mean?

Obin Olson · July 11th, 2004, 11:53 AM

Steve your going to get a really BAD FOV with that 1/2 3.2megapixel chip if you can't do some sort of pixel binning....I hate to say it but we would not want a roi that kills the FOV to a size of a 1/3 inch chip! That will put the image DOF back in Handycam land..not a good thing

Jason Rodriguez · July 11th, 2004, 12:24 PM

BTW,

FYI, if you're using a 2/3" chip, you're going to get around the same DOF as a 16mm frame.

Now unlike video lenses which have to deal with prism optics, 16mm film lenses open up way more, like down to f1.3, and at that f-stop, you're getting around the same DOF as a f2.8-f4 split in 35mm, which is not a lot of DOF at all-that can give you very shallow DOF like you're wanting without the need for a 35mm ground-glass optical-type effect.

For instance, there's a reason that people pay $115K for a set of Zeiss digiprimes. They can basically open up to F1.6 (the maximum f-stop for B4-mount (2/3 inch Sony mount) lenses), and be sharp as a tack whereas many other lenses fall apart at that F-stop on video cameras (again, there's a prism block limiting the maximum aperture of a video camera). With a single-sensor design there is no such problem, hence you can use optics that are very fast, and as a result get that nice out-of-focus DOF you've been wanting, without the need for a ground-glass 35mm converter.

Obin Olson · July 11th, 2004, 12:59 PM

Thank you Jason, that is what I was thinking - now how do you find a f1.3-1.6 lens that is sharp at that f-stop? I will not buy any $115k primes soon :) any ideas ?

Jason Rodriguez · July 11th, 2004, 02:20 PM

Any of the Zeiss superspeeds go to F1.3. Also you have primes from Optimo (a distributer of russian equipment) that also go to F1.3 IIRC.

The only catch is that these are PL mount, but you can find adapters for C-mount to PL mount, and these lenses aren't THAT much, around $2K new for the Optimo's (which is a very good price for good glass). I'm sure you can find used Zeiss superspeeds, etc. on ebay or Mandy.com.

Rob Scott · July 11th, 2004, 03:10 PM

Quote:

Juan M. M. Fiebelkorn wrote:
I'm feeling sad at this moment cause I see the technical part of this thread seems to go nowhere most of the time....

We could always start another thread for detailed technical stuff and try to stay on topic :-)

Quote:

When the moment arrives, I will need some expert coders to optimize it. Anyone here is able for this task?

If it fits in with my own plans, I might be interested. Let me know how your progress is doing.

Quote:

Everybody seems to ask for complete off-the-shelf FPGA projects.The Russian Net-Camera is a working FPGA project...

Quote:

A Xilinx Spartan3 FPGA 1 million gates costs $15

You're absolutely right. FPGA development boards are cheap, so I'm planning to take a good look at that "Russian" project when I'm done with my current project. I think a "box" camera with

An AltaSens chip
FPGA real-time Bayer
FPGA lossy-but-good compression (wavelets?)
Output via Gigabit Ethernet

... would be very cool.

Obin Olson · July 11th, 2004, 03:34 PM

even cooler is all the above WITH onboard SATA hard disks for capture and dsp to take care of viewfinders and playback screens

Jason Rodriguez · July 11th, 2004, 04:13 PM

Quote:

even cooler is all the above WITH onboard SATA hard disks for capture and dsp to take care of viewfinders and playback screens

hmmm . . . <cough>kinetta??<cough>

:-)

Steve Nordhauser · July 11th, 2004, 05:54 PM

The 3.2Mpix Micron is *not* a replacement for the Altasens. It is just a different product that might fit a certain set of needs. Here is where the Micron data lives:
http://www.micron.com/products/imaging/products/datasheets.html

Wayne on Gigabit:
Most of our cameras will end up on gigabit over time. This is the raw sensor data being transmitted. When using an Intel Pro1000 interface card (<$50) and our custom drivers, expect 800+Mbps (that is bits) with any pixel over 8 bits taking 16 bits. We just tested 1920x1080x24fps@10 bits (796Mbps) and got continuous transfer and display with the SI-3300.

Obin on lenses:
For new lenses compatible with 3 micron pixel pitch, we resell Tamron, Schneider and Linos (Rodenstock). Many of the older lenses are also good since 3 microns is somewhere around the film grain on fine grain film. For $50, take a few shots. The manufacturers I listed start at around $850 for a good 25mm lens. I've had people on ebay do the size test for me on lenses before:

You can check this by holding the lens a couple of inches from a piece of paper, threaded side down. Move it up and down until you get a good image. Measure your height to the mounting shoulder. That is the flange distance. The diameter of the circle will also give you the format (1/4" to 1"). Flange distance of 17mm is c mount 12mm is cs mount (no good)

On FPGAs:
I will say again, if someone has a working design, I will consider licensing it and embedding it in a camera. Right now processors are just getting to where they can do lots of good things in real time - see Cineform for an example of what can be done on a $700 PC.

On this thread out of hand:
Anyone want to prune out the good stuff and add it to or create a new wiki? The ideal structure after the discussion dust is over would be a hypertreaded document. There is certainly a wealth of information in Alternative Imaging Methods.

Juan M. M. Fiebelkorn · July 11th, 2004, 06:14 PM

Guys, what we are planning with FPGA is the same and more tha what we can do with DSP :)
I don't quite understand why everybody thinks that FPGA are expensive, that their development and programming is expensive and that they are power hungry......it seems so many lies have been told to poor consumers :).

I VOTE for a new Thread, only technical!!
And with access only to registered users (no need to tell you why ;) )

@Wayne, sorry guy but sometimes you dream to much, like a brainstorming of one guy!!
Don't expect the Via machines to be a DSP laboratory.
What is at hand now is more than enough.I say this cause the easiest way now would be to use one of that C3 machines with a simple $100 FPGA developmente board (PCI interface), program some simple routines on the FPGA (to compress?) and then recording to disk.
This way we use normal C code, combined with Handel-C or SystemC to offload from CPU intensive tasks.We have the IDE interfaces of the motherboard, we use Linux, we program a graphical interface to control the Camera System, we use a LCD touchscreen of 7" diagonal (directly conected to Motherboard's VGA) and we also add some De-Bayer simple algo to show the image realtime on the touchscreen.
This will drain only around 15-20 Watts.

@Steve: I don't understand.Doesn't Elphel's camera work????
isn't it open source?
Can't it be licensed?
Cineform's codec is great,sure!! But, would you use a CMOS camera that needs more than 100 watts and a big and heavy cooling system???
GEthernet is great, what would happen if someone combines the DCT transform that comes FREE with some Xilinx chips and an entropy encoding (maybe RLE ).My last tests show a lossless but slow compression with an average ratio of 12:1 for a lossless DCT+Huffman.
I said this because I've posted some huffman encoding implemented in FPGA....

BTW, could someone explain me why the SI-1280F-CL is never mentioned?

July 11th, 2004, 05:54 PM	#659
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	The 3.2Mpix Micron is not a replacement for the Altasens. It is just a different product that might fit a certain set of needs. Here is where the Micron data lives: http://www.micron.com/products/imaging/products/datasheets.html Wayne on Gigabit: Most of our cameras will end up on gigabit over time. This is the raw sensor data being transmitted. When using an Intel Pro1000 interface card (<$50) and our custom drivers, expect 800+Mbps (that is bits) with any pixel over 8 bits taking 16 bits. We just tested 1920x1080x24fps@10 bits (796Mbps) and got continuous transfer and display with the SI-3300. Obin on lenses: For new lenses compatible with 3 micron pixel pitch, we resell Tamron, Schneider and Linos (Rodenstock). Many of the older lenses are also good since 3 microns is somewhere around the film grain on fine grain film. For $50, take a few shots. The manufacturers I listed start at around $850 for a good 25mm lens. I've had people on ebay do the size test for me on lenses before: You can check this by holding the lens a couple of inches from a piece of paper, threaded side down. Move it up and down until you get a good image. Measure your height to the mounting shoulder. That is the flange distance. The diameter of the circle will also give you the format (1/4" to 1"). Flange distance of 17mm is c mount 12mm is cs mount (no good) On FPGAs: I will say again, if someone has a working design, I will consider licensing it and embedding it in a camera. Right now processors are just getting to where they can do lots of good things in real time - see Cineform for an example of what can be done on a $700 PC. On this thread out of hand: Anyone want to prune out the good stuff and add it to or create a new wiki? The ideal structure after the discussion dust is over would be a hypertreaded document. There is certainly a wealth of information in Alternative Imaging Methods. __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

July 11th, 2004, 10:40 AM	#646
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	I like the look of the Zeiss lens I think I may bid on that one!

July 11th, 2004, 10:42 AM	#647
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	Steve how can I tell the junk on ebay is the right size - next the ebayer seller will not take it back ;)

July 11th, 2004, 11:00 AM	#649
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	Wayne why do I find it hard to understand what you say in your posts? is it just me??

July 11th, 2004, 11:48 AM	#651
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	see what I mean?

July 11th, 2004, 11:53 AM	#652
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	Steve your going to get a really BAD FOV with that 1/2 3.2megapixel chip if you can't do some sort of pixel binning....I hate to say it but we would not want a roi that kills the FOV to a size of a 1/3 inch chip! That will put the image DOF back in Handycam land..not a good thing

July 11th, 2004, 12:24 PM	#653
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	BTW, FYI, if you're using a 2/3" chip, you're going to get around the same DOF as a 16mm frame. Now unlike video lenses which have to deal with prism optics, 16mm film lenses open up way more, like down to f1.3, and at that f-stop, you're getting around the same DOF as a f2.8-f4 split in 35mm, which is not a lot of DOF at all-that can give you very shallow DOF like you're wanting without the need for a 35mm ground-glass optical-type effect. For instance, there's a reason that people pay $115K for a set of Zeiss digiprimes. They can basically open up to F1.6 (the maximum f-stop for B4-mount (2/3 inch Sony mount) lenses), and be sharp as a tack whereas many other lenses fall apart at that F-stop on video cameras (again, there's a prism block limiting the maximum aperture of a video camera). With a single-sensor design there is no such problem, hence you can use optics that are very fast, and as a result get that nice out-of-focus DOF you've been wanting, without the need for a ground-glass 35mm converter.

July 11th, 2004, 12:59 PM	#654
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	Thank you Jason, that is what I was thinking - now how do you find a f1.3-1.6 lens that is sharp at that f-stop? I will not buy any $115k primes soon :) any ideas ?

July 11th, 2004, 02:20 PM	#655
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	Any of the Zeiss superspeeds go to F1.3. Also you have primes from Optimo (a distributer of russian equipment) that also go to F1.3 IIRC. The only catch is that these are PL mount, but you can find adapters for C-mount to PL mount, and these lenses aren't THAT much, around $2K new for the Optimo's (which is a very good price for good glass). I'm sure you can find used Zeiss superspeeds, etc. on ebay or Mandy.com.

July 11th, 2004, 03:34 PM	#657
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	even cooler is all the above WITH onboard SATA hard disks for capture and dsp to take care of viewfinders and playback screens

July 11th, 2004, 06:14 PM	#660
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	Guys, what we are planning with FPGA is the same and more tha what we can do with DSP :) I don't quite understand why everybody thinks that FPGA are expensive, that their development and programming is expensive and that they are power hungry......it seems so many lies have been told to poor consumers :). I VOTE for a new Thread, only technical!! And with access only to registered users (no need to tell you why ;) ) @Wayne, sorry guy but sometimes you dream to much, like a brainstorming of one guy!! Don't expect the Via machines to be a DSP laboratory. What is at hand now is more than enough.I say this cause the easiest way now would be to use one of that C3 machines with a simple $100 FPGA developmente board (PCI interface), program some simple routines on the FPGA (to compress?) and then recording to disk. This way we use normal C code, combined with Handel-C or SystemC to offload from CPU intensive tasks.We have the IDE interfaces of the motherboard, we use Linux, we program a graphical interface to control the Camera System, we use a LCD touchscreen of 7" diagonal (directly conected to Motherboard's VGA) and we also add some De-Bayer simple algo to show the image realtime on the touchscreen. This will drain only around 15-20 Watts. @Steve: I don't understand.Doesn't Elphel's camera work???? isn't it open source? Can't it be licensed? Cineform's codec is great,sure!! But, would you use a CMOS camera that needs more than 100 watts and a big and heavy cooling system??? GEthernet is great, what would happen if someone combines the DCT transform that comes FREE with some Xilinx chips and an entropy encoding (maybe RLE ).My last tests show a lossless but slow compression with an average ratio of 12:1 for a lossless DCT+Huffman. I said this because I've posted some huffman encoding implemented in FPGA.... BTW, could someone explain me why the SI-1280F-CL is never mentioned?