Sumix 2/3" 1920x1080 CMOS

Daniel Lipats · December 26th, 2008, 12:05 PM

Eric,

Your results look great. What is the workflow like with your software? I would love to have a chance to try it out one day.

It's nice to see you on board :)

I have been really busy with work lately and not had much time to work with the Sumix camera. I also got my hands on a 5D Mark II a while back and that has not helped.

This month I received an updated camera and new software.

Sumix has made a lot of improvements to the software. Like Eric mentioned it is now possible to see what you are recording while you record. It automatically adjusts the picture to a preview setting. Also, you no longer have to specify how many frames to record. Just hit Stop or ESC when done.

They have included a high speed shutter mode. The camera shoots at twice the framerate to reduce the rolling shutter distortion and it really makes a difference.

Finally, Sumix has included a raw workflow. It is now possible to record 10 bit raw video. The raw file output can later be loaded into a raw to avi converter where lookup tables, white balance and channels can be adjusted before baking the video to AVI. Its a lot like working with a RED camera now.

The noise has been of some concern for me too. I wonder if it has anything to do with the heat. I had it on for about 1 hour and 30 min the other day and the case got so hot that it was almost too hot to touch.

But I have seen some great results from the Altasens 3562 found here: Modula Classic HD Cam

I am confident that we can get a good clean picture.

Attached is a test shot with the new software and a picture of the camera and lens.

Farhad Towfiq · December 27th, 2008, 12:59 PM

Quote:

Originally Posted by Eric Wu

For the latitude... I'm not very experienced but I'd say it's around 8 stops. It's really S/N ratio when talking about digital camera's latitude. It depends on how much noise you can accept when digging up the darker area. I'm not too happy with the noise so far but I think it's good enough if controlled well. I have some ideas that may further reduce the noise, but I'll finish the main functions first.

Here's how the uncorrected color looks like. It's desaturated and green tinted as other raw output camera like Viper or SI-2K.

Eric,

The sensor in this camera ProCamHD3562 does not correct for FPN as in ProCamHD4562 or other Altasens sensors. Many lower cost sensors, like Micron sensors that we use in our SMX-M73 cameras also use Fixed Pattern noise smoothing. Although the FPN smoothing makes the image looks better at first glance, specially for low light conditions, it is non-reversible and removes the flexibility and control from the post processing phase. True raw is indeed without correction for FPN. FPN must not diminish the latitude (dynamic range) if corrected by calibration and look up table. Of course this is serious post processing as every sensor setting will make the FPN different.
I hope people will create post processing filters for non-linear correction of FPN, based on many example images at various senor setting for each individual sensor. The results then will be unmatched from that of those using cameras with automatic FPN correction. However if this proves to be unpractical for a large segment of our customers, then we must move on to ProCamHD4562.

Farhad

Eric Wu · December 29th, 2008, 04:21 AM

Daniel,

My targeted workflow is very simple:

1.The camera software save the uncompressed 12-bit raw data directly to .DNG sequences.

2.Import the DNG sequence into Adobe After Effects and do the one-light by adjusting the Camera Raw meta-data settings.

3.After that it's pretty much the same with any other workflow. Do 16-bit color grading and save to any standard format like QuickTime, AVI, Cineon, DPX, TIFF for further processing and editing.

The point is to preserve as much color information as possible and to be able to use existing production-proven software.

Farhad,

I think raw is good.
My understanding is that the FPN, under a given sensor setting, is an fixed "offset" value for each pixel (meaning, not caused by different sensitivity of each pixel), right? If so then if I can find the correct value for each pixel, I can get rid of it by simply subtracting it from the raw value, right? If my understanding is correct then I believe I can do it while recording instead of post processing.

The ProCamHD4562 sounds like an nice upgrade to me, though.

Biel Bestue · December 30th, 2008, 07:35 PM

eric what a diference would make if you put a magenta filter infont of that green-ish raw image (kinda like the way they did for Zodiac) http://digitalcontentproducer.com/im...onworkflow.pdf

would it keep a little bit more of latitude while lossing a bit of light? does using a lookup table affects the final latitude? i don't see a change in latitude in the "dog exaple" (both raw and corrected seem to have the same latitude) but still, would it help anything?

Eric Wu · December 31st, 2008, 02:42 AM

I recorded 12-bit raw data and according to Sumix the hardware LUT doesn't have effect on 12-bit output. If LUT is applied it could have impact on the latitude instead of helping it. I don't think there's any more latitude you can gain from the sensor when you record 12-bit raw.

Green-ish color means the sensor is more sensitive to the color green than the other two, which is not necessarily a bad thing because human eyes too are more sensitive to the color green. Adding a magenta filter may gain a little more latitude on red and blue channels and lose a little F-stop. But even if the channels are balanced the color will still look very desaturated. Without proper calibration it's very difficult to get correct looking colors.

The reason for this is that the RGB channels of the electronic cameras come from the red, green, blue filters put in front of the sensor cells, but in the real world it's impossible to make the filters pure red, green, or blue. Each of the 3 channels has some mixture of the other two. It's typical to incorporate a 3x3 matrix transform to calibrate the color. Every consumer camera does that before recording or outputing.

Adobe's DNG format has the built-in capability to store the color calibration matrix as non-destructive meta data. I have come up with a calibration matrix for my SMX12A2C camera matching the color output of my Nikon D70s. I don't currently have a Macbeth color checker board so I shot some random items of various colors. here's the comparison. Please do not view this as resolution comparison of any kind. The SMX shot was not properly focused. It's for color calibration only.

Biel Bestue · December 31st, 2008, 06:31 AM

so what you said is that the filter isn't actually needed but with a proper LUT you get the work done (minus the light loss of the filter)?

Biel Bestue · December 31st, 2008, 07:51 AM

excuse my second post...

what i can see to that correction is that what really changes are the highlights, while the overall seems to keep all the luma is the highlighs where a change can be seen

look at the middle of the image

http://www.dvinfo.net/conf/attachmen...0-cmos-raw.jpg

http://www.dvinfo.net/conf/attachmen...0-cmos-smx.jpg

Biel Bestue · January 4th, 2009, 01:41 PM

what would be the rig i would need in order to shot 3d, two laptops with lan? is there any possiblity to reduce it to one computer?

Eric Wu · January 5th, 2009, 03:27 AM

The color calibration matrix is just a way to get the color close to what human eyes see, as a clean start for further color grading. In this case it's just 9 numbers store in the DNG file and the color transformation is done when loading the files into Adobe's Camera RAW, not while recording. The recorded raw data is untouched so there's nothing lost. If you're familiar with DSLR's raw file processing then you'll see the benefit of using DNG format.

I've finished the basic functions of my own program and am building a portable camera body now.
So far it's capable of:
1920x1080 24p 12-bit uncompressed raw recording.
Realtime preview with histogram, while or while not recording.
Synchronized 2-channel 16-bit 48Khz audio recording with realtime level indicator.
Fix-pattern noise removal.
Manual shutter-speed and gain.
UI for touch-screen LCD.

Some features to be added in the future:
Realtime playback.
Other recording formats such as 1280x720 60p, 1920x1080 30p...
Focusing assistance.
...

Biel Bestue · January 5th, 2009, 09:26 AM

Quote:

Originally Posted by Eric Wu

...and am building a portable camera body now...

what exaclty do you mean by that?

Daniel Lipats · January 5th, 2009, 09:54 AM

Eric,

What kind of hardware are you using to record that signal? Is it a raid 0 setup?

What network card are you using? Mine is giving me a bit of trouble as indicated here: http://dl.getdropbox.com/u/305574/smx.MP4

We have tried a few off the shelf cards but they lead to the same result. We suspect that the < $100 network cards can not keep up with the signal.

Eric Wu · January 7th, 2009, 02:46 AM

Biel,
The SMX12A2C need a PC to run. By camera body I mean a portable PC run on batteries, that can be carried around like a normal camera. I guess a laptop would do, but I want something looks more like a camera.

Daniel,
The hardware I used for testing is
Core 2 Duo E6400
2.0G RAM
nVidia 7600GT display card
Intel PRO/1000 GT net adapter
software RAID 0 with 2x 7200rpm HDD

The 2-HDD RAID 0 is barely fast enough for 1920x1080 24p 12-bit. But I am able to record continuously about 30 seconds even with one ordinary HDD if I use most of the RAM for frame buffer. That's quite enough for most shooting.

Did you get the defected results at all frame rate settings? For me it only happened when the data rate exceed the capacity of Gigabit Ethernet. But when it happens it doesn't look quite like your result. It doesn't happen with 1920x1080 24p 12-bit setting though. I've tried 2 different net adapters, both cheap ones, with similar results.

Daniel Lipats · January 8th, 2009, 11:43 PM

Eric,

Thanks for the hardware list. Is this the Intel PRO/1000 GT adapter you are using? Newegg.com - Intel PWLA8391GT 10/ 100/ 1000Mbps PCI PRO/1000 GT Desktop Adapter 1 x RJ45 - Network Interface Cards

That particular video was shot with the camera set to 10bit, 75mhz and 24fps. The result really depends on the bitrate of the signal. Higher frame rates make it much worse. If that network adapter works then I will get one too.

Any progress in putting together a mobile system?

Daniel Lipats · January 9th, 2009, 08:55 PM

Eric,

Do you have any plans to make the software available for download or even for sale? I am very interested in recording to the raw DNG format.

Eric Wu · January 10th, 2009, 04:18 AM

Daniel,

That's exactly the same Intel card I'm using.

Thank you for the interests. My program still need a lot of work. Currently it's far from ready to be used for serious shooting. You'll have to wait until I feel confident enough about it.

December 26th, 2008, 12:05 PM	#496
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Eric, Your results look great. What is the workflow like with your software? I would love to have a chance to try it out one day. It's nice to see you on board :) I have been really busy with work lately and not had much time to work with the Sumix camera. I also got my hands on a 5D Mark II a while back and that has not helped. This month I received an updated camera and new software. Sumix has made a lot of improvements to the software. Like Eric mentioned it is now possible to see what you are recording while you record. It automatically adjusts the picture to a preview setting. Also, you no longer have to specify how many frames to record. Just hit Stop or ESC when done. They have included a high speed shutter mode. The camera shoots at twice the framerate to reduce the rolling shutter distortion and it really makes a difference. Finally, Sumix has included a raw workflow. It is now possible to record 10 bit raw video. The raw file output can later be loaded into a raw to avi converter where lookup tables, white balance and channels can be adjusted before baking the video to AVI. Its a lot like working with a RED camera now. The noise has been of some concern for me too. I wonder if it has anything to do with the heat. I had it on for about 1 hour and 30 min the other day and the case got so hot that it was almost too hot to touch. But I have seen some great results from the Altasens 3562 found here: Modula Classic HD Cam I am confident that we can get a good clean picture. Attached is a test shot with the new software and a picture of the camera and lens. Attached Thumbnails

December 31st, 2008, 02:42 AM	#500
Eric Wu Regular Crew Join Date: May 2007 Location: Taiwan Posts: 32	I recorded 12-bit raw data and according to Sumix the hardware LUT doesn't have effect on 12-bit output. If LUT is applied it could have impact on the latitude instead of helping it. I don't think there's any more latitude you can gain from the sensor when you record 12-bit raw. Green-ish color means the sensor is more sensitive to the color green than the other two, which is not necessarily a bad thing because human eyes too are more sensitive to the color green. Adding a magenta filter may gain a little more latitude on red and blue channels and lose a little F-stop. But even if the channels are balanced the color will still look very desaturated. Without proper calibration it's very difficult to get correct looking colors. The reason for this is that the RGB channels of the electronic cameras come from the red, green, blue filters put in front of the sensor cells, but in the real world it's impossible to make the filters pure red, green, or blue. Each of the 3 channels has some mixture of the other two. It's typical to incorporate a 3x3 matrix transform to calibrate the color. Every consumer camera does that before recording or outputing. Adobe's DNG format has the built-in capability to store the color calibration matrix as non-destructive meta data. I have come up with a calibration matrix for my SMX12A2C camera matching the color output of my Nikon D70s. I don't currently have a Macbeth color checker board so I shot some random items of various colors. here's the comparison. Please do not view this as resolution comparison of any kind. The SMX shot was not properly focused. It's for color calibration only. Attached Thumbnails

December 29th, 2008, 04:21 AM	#498
Eric Wu Regular Crew Join Date: May 2007 Location: Taiwan Posts: 32	Daniel, My targeted workflow is very simple: 1.The camera software save the uncompressed 12-bit raw data directly to .DNG sequences. 2.Import the DNG sequence into Adobe After Effects and do the one-light by adjusting the Camera Raw meta-data settings. 3.After that it's pretty much the same with any other workflow. Do 16-bit color grading and save to any standard format like QuickTime, AVI, Cineon, DPX, TIFF for further processing and editing. The point is to preserve as much color information as possible and to be able to use existing production-proven software. Farhad, I think raw is good. My understanding is that the FPN, under a given sensor setting, is an fixed "offset" value for each pixel (meaning, not caused by different sensitivity of each pixel), right? If so then if I can find the correct value for each pixel, I can get rid of it by simply subtracting it from the raw value, right? If my understanding is correct then I believe I can do it while recording instead of post processing. The ProCamHD4562 sounds like an nice upgrade to me, though.

December 30th, 2008, 07:35 PM	#499
Biel Bestue Regular Crew Join Date: Jul 2005 Posts: 96	eric what a diference would make if you put a magenta filter infont of that green-ish raw image (kinda like the way they did for Zodiac) http://digitalcontentproducer.com/im...onworkflow.pdf would it keep a little bit more of latitude while lossing a bit of light? does using a lookup table affects the final latitude? i don't see a change in latitude in the "dog exaple" (both raw and corrected seem to have the same latitude) but still, would it help anything?

December 31st, 2008, 06:31 AM	#501
Biel Bestue Regular Crew Join Date: Jul 2005 Posts: 96	so what you said is that the filter isn't actually needed but with a proper LUT you get the work done (minus the light loss of the filter)?

December 31st, 2008, 07:51 AM	#502
Biel Bestue Regular Crew Join Date: Jul 2005 Posts: 96	excuse my second post... what i can see to that correction is that what really changes are the highlights, while the overall seems to keep all the luma is the highlighs where a change can be seen look at the middle of the image http://www.dvinfo.net/conf/attachmen...0-cmos-raw.jpg http://www.dvinfo.net/conf/attachmen...0-cmos-smx.jpg

January 4th, 2009, 01:41 PM	#503
Biel Bestue Regular Crew Join Date: Jul 2005 Posts: 96	what would be the rig i would need in order to shot 3d, two laptops with lan? is there any possiblity to reduce it to one computer?

January 5th, 2009, 03:27 AM	#504
Eric Wu Regular Crew Join Date: May 2007 Location: Taiwan Posts: 32	The color calibration matrix is just a way to get the color close to what human eyes see, as a clean start for further color grading. In this case it's just 9 numbers store in the DNG file and the color transformation is done when loading the files into Adobe's Camera RAW, not while recording. The recorded raw data is untouched so there's nothing lost. If you're familiar with DSLR's raw file processing then you'll see the benefit of using DNG format. I've finished the basic functions of my own program and am building a portable camera body now. So far it's capable of: 1920x1080 24p 12-bit uncompressed raw recording. Realtime preview with histogram, while or while not recording. Synchronized 2-channel 16-bit 48Khz audio recording with realtime level indicator. Fix-pattern noise removal. Manual shutter-speed and gain. UI for touch-screen LCD. Some features to be added in the future: Realtime playback. Other recording formats such as 1280x720 60p, 1920x1080 30p... Focusing assistance. ...

January 5th, 2009, 09:54 AM	#506
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Eric, What kind of hardware are you using to record that signal? Is it a raid 0 setup? What network card are you using? Mine is giving me a bit of trouble as indicated here: http://dl.getdropbox.com/u/305574/smx.MP4 We have tried a few off the shelf cards but they lead to the same result. We suspect that the < $100 network cards can not keep up with the signal.

January 7th, 2009, 02:46 AM	#507
Eric Wu Regular Crew Join Date: May 2007 Location: Taiwan Posts: 32	Biel, The SMX12A2C need a PC to run. By camera body I mean a portable PC run on batteries, that can be carried around like a normal camera. I guess a laptop would do, but I want something looks more like a camera. Daniel, The hardware I used for testing is Core 2 Duo E6400 2.0G RAM nVidia 7600GT display card Intel PRO/1000 GT net adapter software RAID 0 with 2x 7200rpm HDD The 2-HDD RAID 0 is barely fast enough for 1920x1080 24p 12-bit. But I am able to record continuously about 30 seconds even with one ordinary HDD if I use most of the RAM for frame buffer. That's quite enough for most shooting. Did you get the defected results at all frame rate settings? For me it only happened when the data rate exceed the capacity of Gigabit Ethernet. But when it happens it doesn't look quite like your result. It doesn't happen with 1920x1080 24p 12-bit setting though. I've tried 2 different net adapters, both cheap ones, with similar results.

January 8th, 2009, 11:43 PM	#508
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Eric, Thanks for the hardware list. Is this the Intel PRO/1000 GT adapter you are using? Newegg.com - Intel PWLA8391GT 10/ 100/ 1000Mbps PCI PRO/1000 GT Desktop Adapter 1 x RJ45 - Network Interface Cards That particular video was shot with the camera set to 10bit, 75mhz and 24fps. The result really depends on the bitrate of the signal. Higher frame rates make it much worse. If that network adapter works then I will get one too. Any progress in putting together a mobile system?

January 9th, 2009, 08:55 PM	#509
Daniel Lipats Major Player Join Date: Feb 2006 Posts: 260	Eric, Do you have any plans to make the software available for download or even for sale? I am very interested in recording to the raw DNG format.

January 10th, 2009, 04:18 AM	#510
Eric Wu Regular Crew Join Date: May 2007 Location: Taiwan Posts: 32	Daniel, That's exactly the same Intel card I'm using. Thank you for the interests. My program still need a lot of work. Currently it's far from ready to be used for serious shooting. You'll have to wait until I feel confident enough about it.