how much space for a minute at DVinfo.net

Biel Bestue · July 14th, 2009, 03:49 AM

how much space is needed for one minute at 720p (binning mode) @ 25fps @ jp4 format?

what would take for a minute "at 720p (b-mode) @ 25fps @ color format if compressing with dirac (at lossy compression)?

godamnit! there is way two much variables now! :-)

oh! another question, if bixel-binning benefits from greater light sensitivity, how much stops does this gain of light means? half a stop, a stop...?

Sebastian Pichelhofer · July 16th, 2009, 08:04 AM

Quote:

Originally Posted by Biel Bestue

how much space is needed for one minute at 720p (binning mode) @ 25fps @ jp4 format?

what would take for a minute "at 720p (b-mode) @ 25fps @ color format if compressing with dirac (at lossy compression)?

godamnit! there is way two much variables now! :-)

oh! another question, if bixel-binning benefits from greater light sensitivity, how much stops does this gain of light means? half a stop, a stop...?

Indeed too many variables, because you also need to specify the jpeg quality level (above 90% the size increases drastically with every percent point without any visual quality increase anymore)

The maximum write speed to camera internal HDD is around 15 Megabytes per second currently. That means almost 1GB per minute.
Anything lower is also easily possible of course.

I suspect to reach 15MB/s you would need to go to 96 or 97% quality with JP4 in 720p mode.

Biel Bestue · July 17th, 2009, 12:52 PM

is the jp4 raw file form the family of jpeg2000 kind of files? is it open source? anyways at what quality should put the compression of jp4 to reach the one gig-one minute? 50%? 25%? what is the exact adventage of the jp4 if it is a lossy format? no chroma subsampling? is there a any RAW format with lossless compression capability?

Andrey Filippov · July 17th, 2009, 02:13 PM

Hi Biel, details of the JP4 are described in this article -
Elphel camera under the hood: from Verilog to PHP - Linux For Devices Articles - Linux for Devices

It is not lossless, but it is "raw" in the sense that color conversion from Bayer mosaic is delegated to the post-processing. This step when the colors are just interpolated causes artifacts. Other than that - it is just modified JPEG, we also use non-linear conversion from 12 bits of the sensor to 8 bits that are compressed in such a way that it matches the noise performance of the sensor so for each of the 255 output steps the senor pixel noise is higher. Details are also in the LinuxDevices article - How many bits are really needed in the image pixels? - Linux For Devices Articles - Linux for Devices - unfortunately, after the site redesign it lost images and javascript code - I hope they will fix that.

I originally developed JP4 for the document scanning ( Elphel Model 323 camera | Elphel, Inc. ) , it is used now in other applications that benefit from the post-processing. The goal of the cameras design is to preserve virtually all the senor data while using available bandwidth (network, mass storage devices).

Sebastian Pichelhofer · July 22nd, 2009, 11:46 AM

Quote:

Originally Posted by Biel Bestue

oh! another question, if bixel-binning benefits from greater light sensitivity, how much stops does this gain of light means? half a stop, a stop...?

Your question got me to test this.

This is the official description from the sensor datasheet:
When "summing" is set all sampled capacitors will be enabled for column readout, resulting in an effective gain equal to the column bin factor.
When "averaging" is set, column averaging will be done, and there will be no
additional gain related to the column bin factor.

So using "sum" method makes the image brighter and averaging improves signal-to-noise ratio.

You can see the 2 modes in action here: Index of /pictures/binning
(Sorry for blurry wall image from my development camera)

Biel Bestue · July 29th, 2009, 03:24 AM

it seems like a stop, maybe two thirds of a stop, so this can be like a normal gain (like any other camera) and an averaging of photosites (wich i think it's the best option, (so best option is no light gain

is there a diference between gain and "summing" ?

can we get a sample of the hight noise of "summing" and the low-high noise of "averaging"? what sensor is this? and what latitude can this sensor get?

Sebastian Pichelhofer · July 31st, 2009, 02:16 AM

Even though both gain (analog/digital) and pixel binning in summing mode result in a brighter image the 2 methods are something completely different.

Binning effectively increases the size of the photosite, bigger photosite -> more light.

Gain increases a photosites charge value by multiplying it with a certain factor.

July 14th, 2009, 03:49 AM	#1
Biel Bestue Regular Crew Join Date: Jul 2005 Posts: 96	how much space for a minute how much space is needed for one minute at 720p (binning mode) @ 25fps @ jp4 format? what would take for a minute "at 720p (b-mode) @ 25fps @ color format if compressing with dirac (at lossy compression)? godamnit! there is way two much variables now! :-) oh! another question, if bixel-binning benefits from greater light sensitivity, how much stops does this gain of light means? half a stop, a stop...?

July 17th, 2009, 12:52 PM	#3
Biel Bestue Regular Crew Join Date: Jul 2005 Posts: 96	is the jp4 raw file form the family of jpeg2000 kind of files? is it open source? anyways at what quality should put the compression of jp4 to reach the one gig-one minute? 50%? 25%? what is the exact adventage of the jp4 if it is a lossy format? no chroma subsampling? is there a any RAW format with lossless compression capability?

July 17th, 2009, 02:13 PM	#4
Andrey Filippov Major Player Join Date: Apr 2006 Location: Magna, Utah Posts: 215	Hi Biel, details of the JP4 are described in this article - Elphel camera under the hood: from Verilog to PHP - Linux For Devices Articles - Linux for Devices It is not lossless, but it is "raw" in the sense that color conversion from Bayer mosaic is delegated to the post-processing. This step when the colors are just interpolated causes artifacts. Other than that - it is just modified JPEG, we also use non-linear conversion from 12 bits of the sensor to 8 bits that are compressed in such a way that it matches the noise performance of the sensor so for each of the 255 output steps the senor pixel noise is higher. Details are also in the LinuxDevices article - How many bits are really needed in the image pixels? - Linux For Devices Articles - Linux for Devices - unfortunately, after the site redesign it lost images and javascript code - I hope they will fix that. I originally developed JP4 for the document scanning ( Elphel Model 323 camera \| Elphel, Inc. ) , it is used now in other applications that benefit from the post-processing. The goal of the cameras design is to preserve virtually all the senor data while using available bandwidth (network, mass storage devices).

July 29th, 2009, 03:24 AM	#6
Biel Bestue Regular Crew Join Date: Jul 2005 Posts: 96	it seems like a stop, maybe two thirds of a stop, so this can be like a normal gain (like any other camera) and an averaging of photosites (wich i think it's the best option, (so best option is no light gain is there a diference between gain and "summing" ? can we get a sample of the hight noise of "summing" and the low-high noise of "averaging"? what sensor is this? and what latitude can this sensor get?

July 31st, 2009, 02:16 AM	#7
Sebastian Pichelhofer Regular Crew Join Date: Jan 2007 Location: Vienna, Austria Posts: 112	Even though both gain (analog/digital) and pixel binning in summing mode result in a brighter image the 2 methods are something completely different. Binning effectively increases the size of the photosite, bigger photosite -> more light. Gain increases a photosites charge value by multiplying it with a certain factor.