New solution for people wanting a DIY HD camera from Micron and Ambarella

Robin Lobel · April 14th, 2008, 02:07 AM

RedCode RAW is 12:1.

Jose A. Garcia · April 14th, 2008, 02:35 AM

So... If we're looking for a way to control everything, 12:1 is a lot of data lost. Is not bad, of course. You can still control many aspects of your clips in post, but what you have to work is not an unaltered clip (well... even a RAW uncompressed clip is somehow "altered", but you get the idea).

Farhad Towfiq · April 14th, 2008, 02:53 AM

12:1 means it is already cooked. I suggest someone does a simulation, synthesizing a short video of a 3D object with subtle 3D features and some edges. Then apply the RedCode Raw. Then see after the 12:1 compression of the RedCode Raw if visual clues for motion and 3D are not distorted. See if color can be manipulated on demand and does not create artifacts.
12:1 compression means from a 12 bit image raw out of sensor, you just need one bit of information per pixel. It is like super concentrate. Raw concentrate is oxymoron. It is only to surprise.

Nicky Campos · April 14th, 2008, 03:33 PM

Farhad... your 3d camera Idea is interesting, please tell us more or email me directly - nicky@gorillafilms.co.za

Farhad Towfiq · April 14th, 2008, 11:33 PM

We see the 3D camera design from an engineering point of view that will solve many fundamental cost issues and technical issues which present Bayer single sensor or 3 sensor prism cameras can not solve. On the other hand 3D output can produce a new world of possibilities for making fantastic movies either in 2D or 3D.
I tried earlier to explain some merits of our 3D camera design for two reasons, first to present a road map that shows-- it would be senseless to go after DOF adapters or 35mm sensors, and second, posting here is a cheap way to protect us from predatory patenting by other companies.

Bayer pattern will always have problem with aliasing. Aliasing will not show itself in single frame static images, but in video, human vision is confused with detecting false motions. No simple low pass optical filter can solve this problem adequately. The ideal filter is not a theoretical spatial low pass filter with sharp cut off frequency (no tail in frequency domain.) The ideal filter is a box filter with no tail in spatial domain. This is exactly what a pixel does with 100% fillfactor or with microlens and no Bayer filter.

If you design lenses for narrow single color light (narrow light bandwidth) and small individual sensors, a $100 lens can beat a $100k zoom lens for 35mm sensor with respect to quality. For each single small lens you have unlimited depth of focus, but when several of them used together for 3D reconstruction then depth of focus can be chosen almost arbitrarily shallow, but only in post processing.

Sensor quality can also improve when they are designed for narrow bandwidth of light. It is possible to achieve QE close to 90%. On the other side shooting redundant sensors at lower frequencies we can capture very low light crevices and this way increase the low dynamic range and low light sensitivity.

Multiple sensors working with offsetting clock edges in effect can add their frame rates together. from 9 sensors with 30 fps each you can produce a 270 fps video. We will not push frame rate that high, but a factor of 2 or 3 will be adequate when needed. On the other side pixel count will also add. So, from several 1.6 mega pixel sensors we can reconstruct 10 mega pixel or more.

The most beautiful aspect of this camera will be the ability to do harmless compression, a kind of compression that is compatible to human vision without distorting 3D and movement visual clues. When there are fast movements pixel resolution can be sacrifices for more continuity in frame to frame movements. When things are slow then pixel count can increase and show high resolution and high contrast effect.

Sensors will be cheap, optics will be cheap, for computation we will depend on Intel's multi-core technology.

On the uses of such camera I am sure there will be creative forces unleashed.
I can only imagine of few narrow ideas. Image enhancement can be done with respect to depth and velocity of objects. Magnifications of object near and far can be altered to some extend. Point of view of camera can be altered in post to some extend. Two or more videos can be combined together, with using depth information so parts of each can overlay on others. You can even use the output for a 3D display when they are perfected.

Juan M. M. Fiebelkorn · April 15th, 2008, 11:59 PM

So, it seems you are somewhat re-inventing Integral Imaging ?

Nicky Campos · April 18th, 2008, 07:12 PM

Farhad... Im really feeling you, forget 35 DOF - choose/emulate in post! Not only high FPS (without needing so much light for shutter), but finally HDR video can be achieved too? I do believe 3d design can solve many fundamental issues for camera design.

I believe you are referring to what is known as "the correspondence problem"? the main obstacle in image-demoisacing? 3d solves that right?

Im just a young film-maker in Africa but I think about this sort of design all the time. I am convinced that in the future these 3d cameras will take off big time especially with 3d compression (another subject).

Ive got a loreo 3d lens arriving soon which Ill put on my 35 adaptor rig on my HD camera, I also plan on purchasing the SVS software from videre design to construct depth maps. I cant wait to play with 3d channels on live-action footage! - Ill share results soon.

Sorry to drift from the original thread - maybe we need a separate section?

April 14th, 2008, 11:33 PM	#20
Farhad Towfiq Regular Crew Join Date: Sep 2007 Location: California Posts: 63	3D camera We see the 3D camera design from an engineering point of view that will solve many fundamental cost issues and technical issues which present Bayer single sensor or 3 sensor prism cameras can not solve. On the other hand 3D output can produce a new world of possibilities for making fantastic movies either in 2D or 3D. I tried earlier to explain some merits of our 3D camera design for two reasons, first to present a road map that shows-- it would be senseless to go after DOF adapters or 35mm sensors, and second, posting here is a cheap way to protect us from predatory patenting by other companies. Bayer pattern will always have problem with aliasing. Aliasing will not show itself in single frame static images, but in video, human vision is confused with detecting false motions. No simple low pass optical filter can solve this problem adequately. The ideal filter is not a theoretical spatial low pass filter with sharp cut off frequency (no tail in frequency domain.) The ideal filter is a box filter with no tail in spatial domain. This is exactly what a pixel does with 100% fillfactor or with microlens and no Bayer filter. If you design lenses for narrow single color light (narrow light bandwidth) and small individual sensors, a $100 lens can beat a $100k zoom lens for 35mm sensor with respect to quality. For each single small lens you have unlimited depth of focus, but when several of them used together for 3D reconstruction then depth of focus can be chosen almost arbitrarily shallow, but only in post processing. Sensor quality can also improve when they are designed for narrow bandwidth of light. It is possible to achieve QE close to 90%. On the other side shooting redundant sensors at lower frequencies we can capture very low light crevices and this way increase the low dynamic range and low light sensitivity. Multiple sensors working with offsetting clock edges in effect can add their frame rates together. from 9 sensors with 30 fps each you can produce a 270 fps video. We will not push frame rate that high, but a factor of 2 or 3 will be adequate when needed. On the other side pixel count will also add. So, from several 1.6 mega pixel sensors we can reconstruct 10 mega pixel or more. The most beautiful aspect of this camera will be the ability to do harmless compression, a kind of compression that is compatible to human vision without distorting 3D and movement visual clues. When there are fast movements pixel resolution can be sacrifices for more continuity in frame to frame movements. When things are slow then pixel count can increase and show high resolution and high contrast effect. Sensors will be cheap, optics will be cheap, for computation we will depend on Intel's multi-core technology. On the uses of such camera I am sure there will be creative forces unleashed. I can only imagine of few narrow ideas. Image enhancement can be done with respect to depth and velocity of objects. Magnifications of object near and far can be altered to some extend. Point of view of camera can be altered in post to some extend. Two or more videos can be combined together, with using depth information so parts of each can overlay on others. You can even use the output for a 3D display when they are perfected.

April 14th, 2008, 02:07 AM	#16
Robin Lobel Regular Crew Join Date: Mar 2008 Location: Paris, France Posts: 80	RedCode RAW is 12:1.

April 14th, 2008, 02:35 AM	#17
Jose A. Garcia Major Player Join Date: May 2007 Location: Sevilla (Spain) Posts: 439	So... If we're looking for a way to control everything, 12:1 is a lot of data lost. Is not bad, of course. You can still control many aspects of your clips in post, but what you have to work is not an unaltered clip (well... even a RAW uncompressed clip is somehow "altered", but you get the idea).

April 14th, 2008, 02:53 AM	#18
Farhad Towfiq Regular Crew Join Date: Sep 2007 Location: California Posts: 63	12:1 means it is already cooked. I suggest someone does a simulation, synthesizing a short video of a 3D object with subtle 3D features and some edges. Then apply the RedCode Raw. Then see after the 12:1 compression of the RedCode Raw if visual clues for motion and 3D are not distorted. See if color can be manipulated on demand and does not create artifacts. 12:1 compression means from a 12 bit image raw out of sensor, you just need one bit of information per pixel. It is like super concentrate. Raw concentrate is oxymoron. It is only to surprise.

April 14th, 2008, 03:33 PM	#19
Nicky Campos Regular Crew Join Date: Jan 2008 Location: Johannesburg, South Africa Posts: 35	Farhad... your 3d camera Idea is interesting, please tell us more or email me directly - nicky@gorillafilms.co.za

April 15th, 2008, 11:59 PM	#21
Juan M. M. Fiebelkorn Major Player Join Date: Jun 2004 Location: Buenos Aires , Argentina Posts: 444	So, it seems you are somewhat re-inventing Integral Imaging ?

April 18th, 2008, 07:12 PM	#22
Nicky Campos Regular Crew Join Date: Jan 2008 Location: Johannesburg, South Africa Posts: 35	Farhad... Im really feeling you, forget 35 DOF - choose/emulate in post! Not only high FPS (without needing so much light for shutter), but finally HDR video can be achieved too? I do believe 3d design can solve many fundamental issues for camera design. I believe you are referring to what is known as "the correspondence problem"? the main obstacle in image-demoisacing? 3d solves that right? Im just a young film-maker in Africa but I think about this sort of design all the time. I am convinced that in the future these 3d cameras will take off big time especially with 3d compression (another subject). Ive got a loreo 3d lens arriving soon which Ill put on my 35 adaptor rig on my HD camera, I also plan on purchasing the SVS software from videre design to construct depth maps. I cant wait to play with 3d channels on live-action footage! - Ill share results soon. Sorry to drift from the original thread - maybe we need a separate section?