Sumix 2/3" 1920x1080 CMOS

Noah Yuan-Vogel · June 2nd, 2008, 10:12 PM

I'm under the impression that contrast issue you are talking about just has to do with recording and display gamma. you might just be seeing the linear sensitivity to light that any such sensor has... youd need use an LUT to make the images gamma-encoded to make it look correct on most monitors for normal viewing. i remember having the same thing happen on my sumix m73. i think you need gamma encoding of around 0.45 to compensate for the normal 2.2 gamma of most displays. I believe DCI says 1/2.6 encoding and 2.6 display gamma is closer to the human eye so it probably keeps more relevant information and displays more naturally. someone correct me if im wrong. im sure SI's camera just compensates for this automatically as any video camera would. of course these machine-vision cameras dont have these settings built in even though they have the same/similar sensors, just takes a bit more programming.

Daniel Lipats · June 2nd, 2008, 10:19 PM

I can do better than that. Here is the video from which the screen capture was taken.

It is in CineForm RAW format, captured in real time with StreamPix 3. Set to FilmScan1 quality. Captured on a quad core (q6600) system. You will see scanlines, this is because my (cheap) gige network adapter is unable to keep up with the signal.

http://dreamstonestudios.com/persona...o/SMX12A2C.avi

Here is the full resolution HV20 image:
http://dreamstonestudios.com/persona...video/hv20.jpg

Solomon Chase · June 2nd, 2008, 11:31 PM

Quote:

Originally Posted by Daniel Lipats

I can do better than that. Here is the video from which the screen capture was taken.

It is in CineForm RAW format, captured in real time with StreamPix 3. Set to FilmScan1 quality. Captured on a quad core (q6600) system. You will see scanlines, this is because my (cheap) gige network adapter is unable to keep up with the signal.

http://dreamstonestudios.com/persona...o/SMX12A2C.avi

Here is the full resolution HV20 image:
http://dreamstonestudios.com/persona...video/hv20.jpg

Noah is right... If you had the new IR cut filter and had underexposed another stop, these images could have looked identical... here is what I was able to do:

Daniel Lipats · June 3rd, 2008, 12:18 AM

My original reply explained that my camera had already been replaced with a new filter.

However, I'm not sure if there is an even newer version or not.

You may be right about underexposing it a bit more. But its heavily underexposed already. If I had underexposed any more the only part of the table you could make out would of been the highlight. I no longer have the HV20 around to test with, so there is no way to be sure.

The HV20 image contains a lot more data in the shadows already. Dropping exposure more on the Sumix image may have lost it all.

Farhad Towfiq · June 3rd, 2008, 12:53 AM

Daniel,

As I mentioned when we sent you the replacement it was a temporary replacement. The filter you have on the SMX-12A2C is still an absorptive filter that is not quite suitable for Altasens sensor. We use this filter on Micron sensor with good results as Micron has narrower range of colors. For altasens we are preparing interference reflective filters with scratch resistance hafnium coating. The design of the head is also changed so the filter can be screwed in and out in a few seconds with fingers without using a tool.

Daniel Lipats · June 3rd, 2008, 01:02 AM

Farhad,

When can we expect an updated filter and software to be ready? Do you think it could make substantial improvements?

Farhad Towfiq · June 3rd, 2008, 01:21 AM

I check the progress tomorrow and let you know. I expect end of June is reasonable, but you can not count on it as the same person coating the filters is also preparing beamsplitters for interferometers.

Paul Curtis · June 3rd, 2008, 01:56 AM

Quote:

Originally Posted by Daniel Lipats

You may be right about underexposing it a bit more. But its heavily underexposed already. If I had underexposed any more the only part of the table you could make out would of been the highlight. I no longer have the HV20 around to test with, so there is no way to be sure.

I think this highlights (no pun intended :) a fundamental difference with these DIY cameras. The signal you get out is much closer to what the sensor saw and in any grown up camera there is an enormous amount of colour processing that happens before anyone sees the image (not just a simple LUT either). In the DIY world there are no (software) tools that handle this processing - sure you can use AE and get most of the way (using the cineon plug in is quite useful) but most of these cameras have different LUTS for different exposures and also fixing the colours as well. Take Vos was doing a lot of work on this for his Pike camera head and im not sure how he got on in the end?

That's one component that all DIYers would need, perhaps a AE plug in that aids in calibration and demosaicing.

SI must have done this work, along with all the software tools to aid in focusing and controlling the image. It's no mean feat and fully justifies the cost of their solution.

CCDs are easier than CMOS as the raw image out is much closer.

Also in Daniels lens tests (the resolution charts) the particular lenses that were being tested really weren't that good. And whilst some of the machine vision lenses do seem to produce nice images a lot of them are very substandard (afterall they're most CCTV lenses). The fujinons are nice and i never manged to test the schneiders which look good on paper.

Obviously with the 2/3rds size in the summix then you could use digiprimes (as SI demonstrate) despite that fact that they're really designed for prism systems. But then you'll need PL to cmount adaptors and probably a head case redesign. Also who has a set of digiprimes laying around, if you did you'd probably have a good camera too... The alternative is the huge array of 16mm lenses but a lot of these won't cover the sensor fully and in fact really aren't that good resolution wise (i tried cookes and switars).

cheers
paul

Noah Yuan-Vogel · June 3rd, 2008, 08:31 AM

I dont think this should be that much of an issue, I'm certain sumix has the ability to do LUTs in hardware as they did with my M73. Perhaps it just isnt implemented yet, or daniel hasnt gone to the trouble of programming it yet. this isnt stuff that should be done in after effects in a normal workflow.

paul, what makes you say raw images from ccd are easier? im not sure the differences are significant. every sensor out there naturally has a linear response to light. whether the camera automatically corrects for colorspace/gamma etc is up to the manufacturer/programmers.

Farhad Towfiq · June 3rd, 2008, 08:41 AM

Daniel,

By enabling hardware gamma correction you can compress the high end of brightness range and avoid saturation.

The new firmware also allows altering exposure time frame by frame and along with a normal exposure frame producing a short exposure frame also, so variation in high intensity spots can be captured.
All the extra possible software niceties are difficult to implement, publish and support as standard features. But some one with strong programming skills can easily manipulate the flexibility of sensor/firmware to his own advantage.

Daniel Lipats · June 3rd, 2008, 08:59 AM

Farhad,

I have no such options. I have not received any new software, i'm still using the original Sumix software that the camera came with.

Paul Curtis · June 3rd, 2008, 08:59 AM

Quote:

Originally Posted by Noah Yuan-Vogel

I dont think this should be that much of an issue, I'm certain sumix has the ability to do LUTs in hardware as they did with my M73. Perhaps it just isnt implemented yet, or daniel hasnt gone to the trouble of programming it yet. this isnt stuff that should be done in after effects in a normal workflow.

paul, what makes you say raw images from ccd are easier? im not sure the differences are significant. every sensor out there naturally has a linear response to light. whether the camera automatically corrects for colorspace/gamma etc is up to the manufacturer/programmers.

Can you apply different LUTS at different exposures (a 3D LUT?)

The CCD vs CMOS thing is an observation. Every 'raw' image out from cmos i've personally seen looks desaturated and the colour red is usually off (even the red camera images seen very flat). CCD on the other hand seems more faithful to the scene. Now whether this is an issue of pixel size (perhaps), because the CMOS has onboard electronics in each pixel reducing the size compared to a similar CCD. Or whether it's fundementally because of the difference in materials i don't know. If anyone can agree or refute my observations i'd like to hear more and understand why.

cheers
paul

Daniel Lipats · June 3rd, 2008, 09:00 AM

Noah,

The image I originally posted was not white balanced, and no LUT was set. This is because the image was recorded as CineForm RAW. White balance and LUT adjustment is not baked as part of the video, you can adjust this in post. Its only metadata.

The video file I provided is a RAW file with Bayer preserved. This is sensor data without any processing. My understanding is that the only thing permanent about it is exposure. If you like, you can apply LUTs to this file and see what you can do

Daniel Lipats · June 3rd, 2008, 01:20 PM

Farhad Towfiq · June 3rd, 2008, 03:23 PM

Daniel,

For new software please send an email to our support.
I suggest if you decide to continue with this project to keep it as simple as possible. Mechanical integration, lenses, computers, display, power supply, programming, third party software, and endless testing is just too much for a single person. For a special effect team with a reasonable budget this activity makes sense. It is funny that Jose was/is contemplating to start from a bare sensor.

June 3rd, 2008, 12:18 AM	#424
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	My original reply explained that my camera had already been replaced with a new filter. However, I'm not sure if there is an even newer version or not. You may be right about underexposing it a bit more. But its heavily underexposed already. If I had underexposed any more the only part of the table you could make out would of been the highlight. I no longer have the HV20 around to test with, so there is no way to be sure. The HV20 image contains a lot more data in the shadows already. Dropping exposure more on the Sumix image may have lost it all. Last edited by Daniel Lipats; June 3rd, 2008 at 12:51 AM.

June 3rd, 2008, 12:53 AM	#425
Farhad Towfiq Regular Crew Join Date: Sep 2007 Location: California Posts: 63	Filter Daniel, As I mentioned when we sent you the replacement it was a temporary replacement. The filter you have on the SMX-12A2C is still an absorptive filter that is not quite suitable for Altasens sensor. We use this filter on Micron sensor with good results as Micron has narrower range of colors. For altasens we are preparing interference reflective filters with scratch resistance hafnium coating. The design of the head is also changed so the filter can be screwed in and out in a few seconds with fingers without using a tool.

June 3rd, 2008, 08:41 AM	#430
Farhad Towfiq Regular Crew Join Date: Sep 2007 Location: California Posts: 63	Hardware (pre A/D) Gamma correction Daniel, By enabling hardware gamma correction you can compress the high end of brightness range and avoid saturation. The new firmware also allows altering exposure time frame by frame and along with a normal exposure frame producing a short exposure frame also, so variation in high intensity spots can be captured. All the extra possible software niceties are difficult to implement, publish and support as standard features. But some one with strong programming skills can easily manipulate the flexibility of sensor/firmware to his own advantage. Attached Thumbnails

June 3rd, 2008, 09:00 AM	#433
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Noah, The image I originally posted was not white balanced, and no LUT was set. This is because the image was recorded as CineForm RAW. White balance and LUT adjustment is not baked as part of the video, you can adjust this in post. Its only metadata. The video file I provided is a RAW file with Bayer preserved. This is sensor data without any processing. My understanding is that the only thing permanent about it is exposure. If you like, you can apply LUTs to this file and see what you can do Last edited by Daniel Lipats; June 3rd, 2008 at 09:48 AM.

June 3rd, 2008, 01:20 PM	#434
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Here are a few more videos. Most of them are just random tests. Shot with StreamPix 3 using the CineForm RAW codec. Filmscan1 quality, on a Core 2 Quad system. Note that scanlines and any other visible artifacts are due to the cheap network card unable to keep up with the bandwidth. [Please forgive the shake. Camera & touchscreen LCD mounted on a not very stable support.] http://www.dreamstonestudios.com/per...2C/video/1.avi http://www.dreamstonestudios.com/per...2C/video/2.avi http://www.dreamstonestudios.com/per...2C/video/3.avi http://www.dreamstonestudios.com/per...2C/video/5.avi http://www.dreamstonestudios.com/per...2C/video/6.avi http://www.dreamstonestudios.com/per...2C/video/7.avi http://www.dreamstonestudios.com/per...derExposed.avi http://www.dreamstonestudios.com/per...2C/video/8.avi http://www.dreamstonestudios.com/per...2C/video/9.avi http://www.dreamstonestudios.com/per...2C/video/10.avi Last edited by Daniel Lipats; June 3rd, 2008 at 02:46 PM.

June 2nd, 2008, 10:12 PM	#421
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	I'm under the impression that contrast issue you are talking about just has to do with recording and display gamma. you might just be seeing the linear sensitivity to light that any such sensor has... youd need use an LUT to make the images gamma-encoded to make it look correct on most monitors for normal viewing. i remember having the same thing happen on my sumix m73. i think you need gamma encoding of around 0.45 to compensate for the normal 2.2 gamma of most displays. I believe DCI says 1/2.6 encoding and 2.6 display gamma is closer to the human eye so it probably keeps more relevant information and displays more naturally. someone correct me if im wrong. im sure SI's camera just compensates for this automatically as any video camera would. of course these machine-vision cameras dont have these settings built in even though they have the same/similar sensors, just takes a bit more programming.

June 2nd, 2008, 10:19 PM	#422
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	I can do better than that. Here is the video from which the screen capture was taken. It is in CineForm RAW format, captured in real time with StreamPix 3. Set to FilmScan1 quality. Captured on a quad core (q6600) system. You will see scanlines, this is because my (cheap) gige network adapter is unable to keep up with the signal. http://dreamstonestudios.com/persona...o/SMX12A2C.avi Here is the full resolution HV20 image: http://dreamstonestudios.com/persona...video/hv20.jpg

June 3rd, 2008, 01:02 AM	#426
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Farhad, When can we expect an updated filter and software to be ready? Do you think it could make substantial improvements?

June 3rd, 2008, 01:21 AM	#427
Farhad Towfiq Regular Crew Join Date: Sep 2007 Location: California Posts: 63	I check the progress tomorrow and let you know. I expect end of June is reasonable, but you can not count on it as the same person coating the filters is also preparing beamsplitters for interferometers.

June 3rd, 2008, 08:31 AM	#429
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	I dont think this should be that much of an issue, I'm certain sumix has the ability to do LUTs in hardware as they did with my M73. Perhaps it just isnt implemented yet, or daniel hasnt gone to the trouble of programming it yet. this isnt stuff that should be done in after effects in a normal workflow. paul, what makes you say raw images from ccd are easier? im not sure the differences are significant. every sensor out there naturally has a linear response to light. whether the camera automatically corrects for colorspace/gamma etc is up to the manufacturer/programmers.

June 3rd, 2008, 08:59 AM	#431
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Farhad, I have no such options. I have not received any new software, i'm still using the original Sumix software that the camera came with.

June 3rd, 2008, 03:23 PM	#435
Farhad Towfiq Regular Crew Join Date: Sep 2007 Location: California Posts: 63	Daniel, For new software please send an email to our support. I suggest if you decide to continue with this project to keep it as simple as possible. Mechanical integration, lenses, computers, display, power supply, programming, third party software, and endless testing is just too much for a single person. For a special effect team with a reasonable budget this activity makes sense. It is funny that Jose was/is contemplating to start from a bare sensor.