4:4:4 10bit single CMOS HD project

Steve Nordhauser · July 28th, 2004, 07:15 AM

Wayne on 1920HD:
That artsy shot was my boss shooting his wife. We are having troubles with Altasens getting more parts, especially color ones. This may push back a formal release date but I am trying to get more engineering parts for the developers - hardware and software

Juan on 3170 sensitivity:
Two separate issues. First, at a gain of 1, you need a brightly lit conference room type lighting (yeah, I know that is not too numerical) to get a good image. Second, as you increase the gain beyond about 3x, the FPN becomes obnoxious. Sure you can background subtract a reference and it doesn't look too bad, but you lose dynamic range doing that. You can clean things up. Look at this (with a poor Bayer, I know):
http://host196.ipowerweb.com/~silico...%208%20bit.jpg
http://host196.ipowerweb.com/~silico...h%20Photo2.jpg
I want to stress that these are well lit, gain and offset corrected, color tweaked, *NOT* raw.

Obin:
I'm glad you are making headway on the SI-1300 smear. Are you shooting at 24 or 48fps?

Ben:
I'm not sure if I see the advantage of the IBIS-5 in rolling shutter mode - larger pixels and cheap interface I guess. The main interest was in using the global shutter to remove the rolling shutter artifacts.

On log encoding:
There are two ways I can think of doing this - first with a log amplifier before the A/D. This will give you more dynamic range a the expense of less detail at one end of the brightness. I assume (??) that you want more detail i the shadows, not less so you get more steps in the dark areas. We just opted to use an external 12 bitA/D instead of the internal 10 bit to accomplish this on our IBIS-5 camera.

The second way would be to remap the 10 bit data down to 8 bit non-linearly to maintain the 10 bit dynamic range and detail at one end of the illumination, again at the expense of the rest of the range.

Personally, I would see more value in either straight 12 bits or going to a lossless compression if data rate is the issue (as in USB). Maybe I'm missing something here.

Wayne Morellini · July 28th, 2004, 07:42 AM

That BBC HD work is great, I'm watching the "Chicargo" musical stage show scene in AB Fab at the moment.

Steve, I'll leave this to the professionals, but I'm sure there is more. I think it is a matter of how our eyes work (non linear response curve with value to value at different distances (the eyes uses around 128 green values etc, 256 is used because that are not even), so if you can select the values according to the eyes performance curve, you can get the 10bit values to go a lot further. I have a link to the spectacular JVC GY-DV5000 camera, that I was reading last night, it uses a simular technique. I think if it is done right, then you adjust for the highlights and most of the picture range will be right, with crushed blacks (but I don't know how this goes in post editing).

http://www.provis.com.au/products/video/gy_dv5000.htm

Can somebody verify if I'm saying the right thing or not?

Quote from the link above:

Quote:

F13 at 2000 lux
The most sensitive camera ever! (F13 at 2000 lux) assures effortless shooting in extreme low light situations. This powerful feature increases creative flexibility and simplifies lighting requirements.

Newly-developed 12-bit ADC and 24-bit DSP
JVC has developed an advanced 12-bit processing system--previously found only in very high end broadcast cameras. The 12-bit ADC (analog to digital converter) directly inputs data to the DSP (digital signal processor,) thus eliminating any signal degradation that might otherwise arise from analog circuits. In addition, the advanced video processing brings out natural details--even in extremely bright scenes, greatly reduces noise, and provides color accuracy found only in the most expensive field production cameras.

400% wide dynamic range
The DV5000's super fast multi-stream parallel processing DSP creates an ultra-smooth gamma curve calculated using a true log scale algorithm. The result is a dynamic range of over 400% that accurately reproduces fine details and colors in both shadowed and highlighted areas.

1/2-inch industry standard bayonet lens mount

400% of what?

Now that is performance, even at 1/2inch, but how does that compare to the Altsens?

I think, I read of a manufacturing problem with the Altsens constricting supply here, pity.

Thanks

Wayne.

Rob Scott · July 28th, 2004, 07:45 AM

Quote:

Wayne Morellini wrote:
Rob S what make of PC do you have, ussually they just exchange main boards to repair.

It's a homebuilt, using an ASUS motherboard. Unfortunately, ASUS does not have a "ship replacement first" option -- I have to ship the motherboard in, they fix/replace, then ship it back. Grrr.

Quote:

Jason Keenan wrote:
Ok, another stupid suggestion. Is there anything in Cinepaint AKA FilmGimp that might be useful?

Very possibly. The plan right now is to make the "Convert" software (the offline, non-real-time portion) available under the GPL, which means we could incorporate code from projects like Cinepaint.

Also, we plan to support file formats that Cinepaint can use directly -- meaning, you could use Cinepaint for color correction, gamma correction, etc. (Assuming that it supports those features -- haven't used it much myself.)

Jason Rodriguez · July 28th, 2004, 08:17 AM

The idea behind Log encoding is that the human response to light is logarithmic, meaning we see more differentiation at the bottom of the scale (darks), than in the highlights. So loss of information in the highlights isn't a problem, and you're packing the bits more effectively. Also log encoding has a nice curve into the highlights, compressing them nicely rather than the harsh clip that happens with linear encoders when they hit 100%. So when you convert from log to linear, you can define a "soft-clip" that allows the highlights to gently roll off-log encoding according to the Cineon spec allows for super-whites, placing 90% white at 695 (I think), and then devoting the rest of the scale above that point to the 1024 cut-off to super-whites. This added information in the highlights helps to make a nice print, since again, there's information in the highlights that makes it to the film negative without any harsh clipping.

Aaron Shaw · July 28th, 2004, 09:18 AM

>>so??
Is it overexposed??
I think we're going the right way now with all this new discoverings and tests...I'm so happy :)<<

Actually I think the image is underexposed.... I was looking at it with the histogram and the blacks are seriously compressed/clipped. I checked out all the other images as well - same problem.

Why is it so necessary to expose for the highlights? Is it not possible to get good blacks without blowing out the image? I admit I'm somewhat lost...

Obin Olson · July 28th, 2004, 09:47 AM

the problem with the 1300 is yes its got good dynamic range but you can't use it when you have a hotspot in your image...you have to crush the blacks enough to "hide" the smear doing so cuts down the range alot...if your not shooting with a hotspot it works GREAT

Jason Rodriguez · July 28th, 2004, 09:54 AM

Quote:

Actually I think the image is underexposed

These images are un-gamma corrected linear images, so yes, they actually are overexposed a little bit (although the last couple that Obin posted with the girl are pretty good, still might be bright, but not like the others from before).

Les Dit,

I was thinking that Steve might know the ISO.

Another thing you could do is check to see what F-stop gives a grey card lit with 2000lux a 50% grey value (after gamma correction) on screen. From there you could back-track to an ISO figure.

Ben Syverson · July 28th, 2004, 11:03 AM

@JasonKeenan: Is there anything in Cinepaint AKA FilmGimp that might be useful?

No. Cinepaint is a frame-by-frame retouching tool, not an editing tool. It's literally mostly used for "dustbusting," or painting dust out of film scans. It's also an extremely early beta (I would call it an alpha), so while certain features like the "open" menu item and the "brush" tool are usable because they're the main features, pretty much everything else is some form of broken.

But I lurk on the Cinepaint developer list to get an idea what they're doing. 0.19 should be released fairly soon.

@Steve: I'm not sure if I see the advantage of the IBIS-5 in rolling shutter mode - larger pixels and cheap interface I guess.

Yeah, not much of an advantage, I guess. No wait, it is!

I could never shoot with a 1/2" sensor, because you can't get quality lenses to go wide enough. I ordered a 10mm prime lens designed for 16mm, and even that lens with 2/3" format is not that wide (in 16mm, it's super wide). To get wide-angle in 1/2" you're basically stuck with industrial video lenses (of extremely questionable quality) or super-expensive 16mm primes.

And the cheap interface is a huge, huge thing, especially for someone such as myself who wants to shoot outside. Also, Sumix is commited to similar interfaces for future products. Much like the SI-1300 is "practice" for Obin before the Altasens comes out, so is the SMX-150c for me.

Just curious, what interface(s) are you putting on the Altasens design?

- ben

Aaron Shaw · July 28th, 2004, 11:17 AM

Obin, how do you go about eleminating hot spots? Is this a consistent problem?

Adjusting the gamma shouldn't be a large problem. I'm just not sure I fully understand how you can get crushed blacks from an overexposed image. Is it because you have to pull the highlights down so much?

Steve Nordhauser · July 28th, 2004, 12:22 PM

Ben:
The altasens will be first on camera link since its top end is about 300MB/sec. We will then do a gigE version but should include at least packing if not compression.

There are Schneider Cinegon 8mm lenses at 2/3". Not cheap but none of the good glass is unless you buy it used.

I will be curious how you fare on the USB 2.0 because we have them too and have written them off for cinematography.

Ben Syverson · July 28th, 2004, 12:29 PM

Steve,

With 10bit transfers (and possibly some compression? it's not clear) USB2 should be fine for 720p at 24mhz. It's FW800 that I'm really looking forward to..

Funny you should mention Scneider Cinegon -- the 8mm was a little out of my league, but I have a used (well, old -- I don't think it was ever used) 10mm Cinegon on the way. Supposedly it covers all the way to Super16, so maybe it's an older model...

Good to hear about GigE -- will the camera get its power over ethernet as well?

- ben

Obin Olson · July 28th, 2004, 01:40 PM

hotspots are jsut areas in the image that have above 100% exposure...it's what you shoot not a "spot" on the image

Eric Gorski · July 28th, 2004, 02:09 PM

who's going to have the first working altasens camera?? and when? and for how much? any guesses?

Obin Olson · July 28th, 2004, 02:28 PM

check out the dynamic range in this image! WOW NO lights on inside at all and you can still see the window sill if you push it...but see the streaking? that is what I am talking about...take the gamma all the way to 2.6 or so and you will see what I mean...

http://obin.weet.us/16bit_color_tiff.tif

Rai Orz · July 28th, 2004, 02:51 PM

Steve, i wanted today order a SI-1300-Set, but the rolling shutter problem will not go out of my head. You know, we need the camera for a movie project. Shoot with 48fps, but only use every second frame is also not like a movie camera. Okay, in a movie camera (with 24fps) the exposure time is about 1/50 sec., but not the exposure time is the problem. It is the slow rolles shuttet. A mechanical shutter in a movie camera move in a different way and is faster. The shutter diameter is 3-5 times larger then the frame size. So from a whole one turn it need only a angle of round 15 degree to close (or open) the picture window. This time is only a 1/24 of one frame (one turn=360degree, so 15degree = 1/24).
1/24 of one frame, not 1/24sec like the SI-1300. A movie camera shutter rolls (at 24fps) in 1/576 sec. This is 24 times faster! This is a enormous difference and i think definitely rolling shutters are not usefully to shoot a movie with it.

@Ben, you have no email here, can you contacted my, please

July 28th, 2004, 07:15 AM	#1171
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	Wayne on 1920HD: That artsy shot was my boss shooting his wife. We are having troubles with Altasens getting more parts, especially color ones. This may push back a formal release date but I am trying to get more engineering parts for the developers - hardware and software Juan on 3170 sensitivity: Two separate issues. First, at a gain of 1, you need a brightly lit conference room type lighting (yeah, I know that is not too numerical) to get a good image. Second, as you increase the gain beyond about 3x, the FPN becomes obnoxious. Sure you can background subtract a reference and it doesn't look too bad, but you lose dynamic range doing that. You can clean things up. Look at this (with a poor Bayer, I know): http://host196.ipowerweb.com/~silico...%208%20bit.jpg http://host196.ipowerweb.com/~silico...h%20Photo2.jpg I want to stress that these are well lit, gain and offset corrected, color tweaked, NOT raw. Obin: I'm glad you are making headway on the SI-1300 smear. Are you shooting at 24 or 48fps? Ben: I'm not sure if I see the advantage of the IBIS-5 in rolling shutter mode - larger pixels and cheap interface I guess. The main interest was in using the global shutter to remove the rolling shutter artifacts. On log encoding: There are two ways I can think of doing this - first with a log amplifier before the A/D. This will give you more dynamic range a the expense of less detail at one end of the brightness. I assume (??) that you want more detail i the shadows, not less so you get more steps in the dark areas. We just opted to use an external 12 bitA/D instead of the internal 10 bit to accomplish this on our IBIS-5 camera. The second way would be to remap the 10 bit data down to 8 bit non-linearly to maintain the 10 bit dynamic range and detail at one end of the illumination, again at the expense of the rest of the range. Personally, I would see more value in either straight 12 bits or going to a lossless compression if data rate is the issue (as in USB). Maybe I'm missing something here. __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

July 28th, 2004, 12:22 PM	#1180
Steve Nordhauser Silicon Imaging, Inc. Join Date: May 2004 Location: Troy, NY USA Posts: 325	Ben: The altasens will be first on camera link since its top end is about 300MB/sec. We will then do a gigE version but should include at least packing if not compression. There are Schneider Cinegon 8mm lenses at 2/3". Not cheap but none of the good glass is unless you buy it used. I will be curious how you fare on the USB 2.0 because we have them too and have written them off for cinematography. __________________ Silicon Imaging, Inc. We see the Light! http://www.siliconimaging.com

July 28th, 2004, 08:17 AM	#1174
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	The idea behind Log encoding is that the human response to light is logarithmic, meaning we see more differentiation at the bottom of the scale (darks), than in the highlights. So loss of information in the highlights isn't a problem, and you're packing the bits more effectively. Also log encoding has a nice curve into the highlights, compressing them nicely rather than the harsh clip that happens with linear encoders when they hit 100%. So when you convert from log to linear, you can define a "soft-clip" that allows the highlights to gently roll off-log encoding according to the Cineon spec allows for super-whites, placing 90% white at 695 (I think), and then devoting the rest of the scale above that point to the 1024 cut-off to super-whites. This added information in the highlights helps to make a nice print, since again, there's information in the highlights that makes it to the film negative without any harsh clipping.

July 28th, 2004, 09:18 AM	#1175
Aaron Shaw Major Player Join Date: Jul 2004 Posts: 587	>>so?? Is it overexposed?? I think we're going the right way now with all this new discoverings and tests...I'm so happy :)<< Actually I think the image is underexposed.... I was looking at it with the histogram and the blacks are seriously compressed/clipped. I checked out all the other images as well - same problem. Why is it so necessary to expose for the highlights? Is it not possible to get good blacks without blowing out the image? I admit I'm somewhat lost...

July 28th, 2004, 09:47 AM	#1176
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	the problem with the 1300 is yes its got good dynamic range but you can't use it when you have a hotspot in your image...you have to crush the blacks enough to "hide" the smear doing so cuts down the range alot...if your not shooting with a hotspot it works GREAT

July 28th, 2004, 11:03 AM	#1178
Ben Syverson Major Player Join Date: Mar 2004 Location: chicago Posts: 434	@JasonKeenan: Is there anything in Cinepaint AKA FilmGimp that might be useful? No. Cinepaint is a frame-by-frame retouching tool, not an editing tool. It's literally mostly used for "dustbusting," or painting dust out of film scans. It's also an extremely early beta (I would call it an alpha), so while certain features like the "open" menu item and the "brush" tool are usable because they're the main features, pretty much everything else is some form of broken. But I lurk on the Cinepaint developer list to get an idea what they're doing. 0.19 should be released fairly soon. @Steve: I'm not sure if I see the advantage of the IBIS-5 in rolling shutter mode - larger pixels and cheap interface I guess. Yeah, not much of an advantage, I guess. No wait, it is! I could never shoot with a 1/2" sensor, because you can't get quality lenses to go wide enough. I ordered a 10mm prime lens designed for 16mm, and even that lens with 2/3" format is not that wide (in 16mm, it's super wide). To get wide-angle in 1/2" you're basically stuck with industrial video lenses (of extremely questionable quality) or super-expensive 16mm primes. And the cheap interface is a huge, huge thing, especially for someone such as myself who wants to shoot outside. Also, Sumix is commited to similar interfaces for future products. Much like the SI-1300 is "practice" for Obin before the Altasens comes out, so is the SMX-150c for me. Just curious, what interface(s) are you putting on the Altasens design? - ben

July 28th, 2004, 11:17 AM	#1179
Aaron Shaw Major Player Join Date: Jul 2004 Posts: 587	Obin, how do you go about eleminating hot spots? Is this a consistent problem? Adjusting the gamma shouldn't be a large problem. I'm just not sure I fully understand how you can get crushed blacks from an overexposed image. Is it because you have to pull the highlights down so much?

July 28th, 2004, 12:29 PM	#1181
Ben Syverson Major Player Join Date: Mar 2004 Location: chicago Posts: 434	Steve, With 10bit transfers (and possibly some compression? it's not clear) USB2 should be fine for 720p at 24mhz. It's FW800 that I'm really looking forward to.. Funny you should mention Scneider Cinegon -- the 8mm was a little out of my league, but I have a used (well, old -- I don't think it was ever used) 10mm Cinegon on the way. Supposedly it covers all the way to Super16, so maybe it's an older model... Good to hear about GigE -- will the camera get its power over ethernet as well? - ben

July 28th, 2004, 01:40 PM	#1182
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	hotspots are jsut areas in the image that have above 100% exposure...it's what you shoot not a "spot" on the image

July 28th, 2004, 02:09 PM	#1183
Eric Gorski Regular Crew Join Date: Jun 2004 Location: Western Oregon Posts: 138	who's going to have the first working altasens camera?? and when? and for how much? any guesses?

July 28th, 2004, 02:28 PM	#1184
Obin Olson Trustee Join Date: Jan 2003 Location: Wilmington NC Posts: 1,414	check out the dynamic range in this image! WOW NO lights on inside at all and you can still see the window sill if you push it...but see the streaking? that is what I am talking about...take the gamma all the way to 2.6 or so and you will see what I mean... http://obin.weet.us/16bit_color_tiff.tif

July 28th, 2004, 02:51 PM	#1185
Rai Orz Regular Crew Join Date: Jun 2004 Location: Germany Posts: 136	Steve, i wanted today order a SI-1300-Set, but the rolling shutter problem will not go out of my head. You know, we need the camera for a movie project. Shoot with 48fps, but only use every second frame is also not like a movie camera. Okay, in a movie camera (with 24fps) the exposure time is about 1/50 sec., but not the exposure time is the problem. It is the slow rolles shuttet. A mechanical shutter in a movie camera move in a different way and is faster. The shutter diameter is 3-5 times larger then the frame size. So from a whole one turn it need only a angle of round 15 degree to close (or open) the picture window. This time is only a 1/24 of one frame (one turn=360degree, so 15degree = 1/24). 1/24 of one frame, not 1/24sec like the SI-1300. A movie camera shutter rolls (at 24fps) in 1/576 sec. This is 24 times faster! This is a enormous difference and i think definitely rolling shutters are not usefully to shoot a movie with it. @Ben, you have no email here, can you contacted my, please