HV20 Latitude and exposure details at DVinfo.net

Noah Yuan-Vogel · May 8th, 2007, 01:29 AM

I just did a relatively crude test of the HV20's latitude, and for fun superimposed it on another latitude chart relating to much higher-end HD cameras (originating here http://www.cinematography.net/hdcamtests/HDcurves.htm). Here's my version.

Note that this was done in using still images saved to minisd with these settings:

24p HDV
Cine mode
whitebalance daylight
sharpness -1
contrast -1
color gain +1

The test shot was bracketed over 16 stops, and is subject to any inaccuracies in the exposure selection system in the hv20. It is assumed that each click of +-1 exposure is 1/4 stop. Measurements were taken every other click, aka every .5 stops. I expect it is best to allow for +-0.5 stops error (which is conservative as long as the exposure meter in the camera is consistent). Fstop and shutter readings were checked at each measurement so exposure steps appear to be consistent. There are 10 +-1 exposure steps (probably 2.5stops), however, that are not accounted for, they may be gain. The sequence of aperture and shutter changes used to get the various exposures are something I posted on in another forum: http://www.dvinfo.net/conf/showpost....0&postcount=21 (some parts need correction)

Chris Barcellos · May 8th, 2007, 08:03 AM

Noah:

So what does it all mean ??

Mike Thomann · May 8th, 2007, 08:25 AM

...what model is the Canon camera with the cyan line?

Noah Yuan-Vogel · May 8th, 2007, 08:50 AM

mike:
I assume it is the XL-H1

chris:
What does it mean? maybe nothing? But if my methods are comparable with those used to create the original graph, it might mean the HV20 set up in the right way could be comparable in terms of latitude to a much fancier camera. I was pretty impressed with the curve for the camera, it is fairly linear (relative to exponentially increasing exposure values) and the slope is not as steep as it could be, covering a fair range of exposure stops (thats a good thing). That says nothing about noise or iso equivalent ratings. Those will probably be the next thing to test.

Mike Thomann · May 8th, 2007, 10:46 AM

Any input as to how this might be used to determine how many stops of latitude the HV20 has?

Bruce Allen · May 8th, 2007, 11:42 AM

Awesome! I have an HV20 on order and am going to test it with a Stouffer chart - look forward to seeing how my results match yours.

Bruce Allen
www.boacinema.com

Noah Yuan-Vogel · May 8th, 2007, 05:02 PM

I'm no expert but it looks like something like 7.5 stops +-0.5 stop depending on how noise looks at the bottom. I think the curve should probably be shifted to the left since it certainly doesnt have 5 stops overexposure latitude and definitely has more than 2.5 under (who determines 0 exposure anyway? does it really matter after its all been graded?). What is important is the distance between the last low value and first high value. I just put the curve where it fit based on my starting and ending data. Also note this is a graph for all the other cameras when they are rated at ISO 320, which i am sure the HV20 was not rated at 320 for this test, but most of those cameras are supposed to be able to do 320 and still get a pretty optimal image, so i wouldnt necessary consider that too much to their detriment.

Ali Husain · May 8th, 2007, 05:38 PM

thanks for doing that test noah!

now if we can only get someone to use the imatest software to get something like (scroll down to "Dynamic Range Analysis):
http://www.imaging-resource.com/PROD...D40IMATEST.HTM

this all depends of course on noise levels. there's a good discussion here:

http://www.normankoren.com/digital_tonality.html

any DSLR tech people have that software?

Ali Husain · May 8th, 2007, 05:46 PM

Quote:

Originally Posted by Chris Barcellos

Noah:

So what does it all mean ??

well the way the hv20 data is presented, it means that the $950 hv20 has better lattitude than every motion picture digital camera except the arri d-20. lol.

okay so probably these datapoints were taken using uncalibrated noise criteria, but i think this is an important comparison: the technology to produce non-stratosphericaly-priced sensors and accompanying electronics for cinema-level capture is JUST ABOUT here.

Ivan Pin · May 9th, 2007, 12:12 AM

There is a russian review HV20 vs HV10: http://www.videozona.ru/video_tests/hv20/

It contains Imatest comparision - measurement of dynamic range, resolution MTF50, color accuracy, ...

A "fun" mashine translated version of the article: http://babelfish.altavista.com/babel...ests%2Fhv20%2F

Ali Husain · May 9th, 2007, 02:57 AM

Quote:

Originally Posted by Ivan Pin

There is a russian review HV20 vs HV10: http://www.videozona.ru/video_tests/hv20/

It contains Imatest comparision - measurement of dynamic range, resolution MTF50, color accuracy, ...

A "fun" mashine translated version of the article: http://babelfish.altavista.com/babel...ests%2Fhv20%2F

wow. tha'ts probably the single most information-full link i've seen posted on this forum. if you go to the end there is this one with more details:

http://babelfish.altavista.com/babel...s%2fImatest%2f

so a summary of the dynamic range result in f-stops vs. image quality is:
9.57 (low), 9.36 (med), 8.64(med-high), 6.13 (high)

which looks pretty good. it matches up to the graph noah sent which i read as showing a best case of a bit over 9-stops.

according to the other link i sent: http://www.imaging-resource.com/PROD...D40IMATEST.HTM it's about the same dynamic range as the canon eos 400 dslr. i couldn't dig up his settings on the test though. can anyone who can read russian tell me what they were? cinemode & contrast minus 1, i believe gives you the greatest range.

also look at this:

color accuracy with lots of light: http://www.videozona.ru/video_tests/...colorerror.png

color accuracy in low-light:
http://www.videozona.ru/video_tests/...colorerror.png

ouch. that agrees with the desaturated look i see when it gets dark.

Fergus Anderson · May 9th, 2007, 05:55 AM

Quote:

Originally Posted by Ali Husain

wow. tha'ts probably the single most information-full link i've seen posted on this forum. if you go to the end there is this one with more details:

http://babelfish.altavista.com/babel...s%2fImatest%2f

so a summary of the dynamic range result in f-stops vs. image quality is:
9.57 (low), 9.36 (med), 8.64(med-high), 6.13 (high)

which looks pretty good. it matches up to the graph noah sent which i read as showing a best case of a bit over 9-stops.

according to the other link i sent: http://www.imaging-resource.com/PROD...D40IMATEST.HTM it's about the same dynamic range as the canon eos 400 dslr. i couldn't dig up his settings on the test though. can anyone who can read russian tell me what they were? cinemode & contrast minus 1, i believe gives you the greatest range.

also look at this:

color accuracy with lots of light: http://www.videozona.ru/video_tests/...colorerror.png

color accuracy in low-light:
http://www.videozona.ru/video_tests/...colorerror.png

ouch. that agrees with the desaturated look i see when it gets dark.

sorry for my ignorance but can someone explain why lowering the contrast increases dynamic range? So the best option is cine with contrat -1 and brightness at 0?

Thanks

Noah Yuan-Vogel · May 9th, 2007, 08:47 AM

my explanation would be its because part of the problem in dynamic range in digital cameras naitively see light variations linearly but light increases exponentially. now if the data came straight from the sensor in the 8-bits (0-255 whole number values), it would be impossible for it to show more than 7-8 stops of light because the brightest 1-stop range of light in the image would be stored in the values 127-255 (half the possible values), the next 63-127, and so on down to 0-1 (6 stops later). so if there were any image data darker than that, it would get rounded to 0 or 1, making it pretty much unrecoverable. fortunately the data from a light sensor usually comes in in 10 or 12 or 14 bit form (allowing 1024, 4096, or 16384 possible values) so if we assume a 12bit image sensor, all the infinite dark stops of light in image that would have all been recorded into the range 0-1 and essentially crushed is now being recorded into 0-15 which allows for several stops of light in that dark detail to be differentiated, giving 4 more potential stops of light at 7-15, 3-7, 1-3 and 0-1. this data is then processed into 8bit (0-255) data to appear more like the curve i posted above and preserve that dark detail. lowering the contrast changes that processing to try to pull more data out of the bright and dark ends (where detail is normally lost)

Mike Thomann · May 9th, 2007, 09:50 AM

This is a very informative thread! Thanks!

Noah Yuan-Vogel · May 9th, 2007, 11:50 AM

i hope my post makes sense. anyway, also, as I understand it, brightness just affects autoexposure. I would probably put brightness at -1 to avoid clipping highlights, but in general i wouldnt use autoexposure much so it wouldnt make much difference. It won't affect latitude.

May 8th, 2007, 01:29 AM	#1
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	HV20 Latitude and exposure details I just did a relatively crude test of the HV20's latitude, and for fun superimposed it on another latitude chart relating to much higher-end HD cameras (originating here http://www.cinematography.net/hdcamtests/HDcurves.htm). Here's my version. Note that this was done in using still images saved to minisd with these settings: 24p HDV Cine mode whitebalance daylight sharpness -1 contrast -1 color gain +1 The test shot was bracketed over 16 stops, and is subject to any inaccuracies in the exposure selection system in the hv20. It is assumed that each click of +-1 exposure is 1/4 stop. Measurements were taken every other click, aka every .5 stops. I expect it is best to allow for +-0.5 stops error (which is conservative as long as the exposure meter in the camera is consistent). Fstop and shutter readings were checked at each measurement so exposure steps appear to be consistent. There are 10 +-1 exposure steps (probably 2.5stops), however, that are not accounted for, they may be gain. The sequence of aperture and shutter changes used to get the various exposures are something I posted on in another forum: http://www.dvinfo.net/conf/showpost....0&postcount=21 (some parts need correction) Attached Thumbnails

May 8th, 2007, 08:03 AM	#2
Chris Barcellos Inner Circle Join Date: Nov 2005 Location: Elk Grove CA Posts: 6,838	Noah: So what does it all mean ?? __________________ Chris J. Barcellos

May 8th, 2007, 08:25 AM	#3
Mike Thomann Regular Crew Join Date: May 2007 Location: Amherst, NY Posts: 100	...what model is the Canon camera with the cyan line?

May 8th, 2007, 08:50 AM	#4
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	mike: I assume it is the XL-H1 chris: What does it mean? maybe nothing? But if my methods are comparable with those used to create the original graph, it might mean the HV20 set up in the right way could be comparable in terms of latitude to a much fancier camera. I was pretty impressed with the curve for the camera, it is fairly linear (relative to exponentially increasing exposure values) and the slope is not as steep as it could be, covering a fair range of exposure stops (thats a good thing). That says nothing about noise or iso equivalent ratings. Those will probably be the next thing to test.

May 8th, 2007, 10:46 AM	#5
Mike Thomann Regular Crew Join Date: May 2007 Location: Amherst, NY Posts: 100	Any input as to how this might be used to determine how many stops of latitude the HV20 has?

May 8th, 2007, 11:42 AM	#6
Bruce Allen Regular Crew Join Date: Jun 2006 Location: Los Angeles, CA Posts: 43	Awesome! I have an HV20 on order and am going to test it with a Stouffer chart - look forward to seeing how my results match yours. Bruce Allen www.boacinema.com

May 8th, 2007, 05:02 PM	#7
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	I'm no expert but it looks like something like 7.5 stops +-0.5 stop depending on how noise looks at the bottom. I think the curve should probably be shifted to the left since it certainly doesnt have 5 stops overexposure latitude and definitely has more than 2.5 under (who determines 0 exposure anyway? does it really matter after its all been graded?). What is important is the distance between the last low value and first high value. I just put the curve where it fit based on my starting and ending data. Also note this is a graph for all the other cameras when they are rated at ISO 320, which i am sure the HV20 was not rated at 320 for this test, but most of those cameras are supposed to be able to do 320 and still get a pretty optimal image, so i wouldnt necessary consider that too much to their detriment.

May 8th, 2007, 05:38 PM	#8
Ali Husain Regular Crew Join Date: Oct 2005 Location: San Jose, CA Posts: 141	thanks for doing that test noah! now if we can only get someone to use the imatest software to get something like (scroll down to "Dynamic Range Analysis): http://www.imaging-resource.com/PROD...D40IMATEST.HTM this all depends of course on noise levels. there's a good discussion here: http://www.normankoren.com/digital_tonality.html any DSLR tech people have that software?

May 9th, 2007, 12:12 AM	#10
Ivan Pin Regular Crew Join Date: Jan 2007 Location: Novosibirsk, Russia Posts: 46	There is a russian review HV20 vs HV10: http://www.videozona.ru/video_tests/hv20/ It contains Imatest comparision - measurement of dynamic range, resolution MTF50, color accuracy, ... A "fun" mashine translated version of the article: http://babelfish.altavista.com/babel...ests%2Fhv20%2F

May 9th, 2007, 08:47 AM	#13
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	my explanation would be its because part of the problem in dynamic range in digital cameras naitively see light variations linearly but light increases exponentially. now if the data came straight from the sensor in the 8-bits (0-255 whole number values), it would be impossible for it to show more than 7-8 stops of light because the brightest 1-stop range of light in the image would be stored in the values 127-255 (half the possible values), the next 63-127, and so on down to 0-1 (6 stops later). so if there were any image data darker than that, it would get rounded to 0 or 1, making it pretty much unrecoverable. fortunately the data from a light sensor usually comes in in 10 or 12 or 14 bit form (allowing 1024, 4096, or 16384 possible values) so if we assume a 12bit image sensor, all the infinite dark stops of light in image that would have all been recorded into the range 0-1 and essentially crushed is now being recorded into 0-15 which allows for several stops of light in that dark detail to be differentiated, giving 4 more potential stops of light at 7-15, 3-7, 1-3 and 0-1. this data is then processed into 8bit (0-255) data to appear more like the curve i posted above and preserve that dark detail. lowering the contrast changes that processing to try to pull more data out of the bright and dark ends (where detail is normally lost)

May 9th, 2007, 09:50 AM	#14
Mike Thomann Regular Crew Join Date: May 2007 Location: Amherst, NY Posts: 100	This is a very informative thread! Thanks!

May 9th, 2007, 11:50 AM	#15
Noah Yuan-Vogel Major Player Join Date: Dec 2004 Location: New York City Posts: 613	i hope my post makes sense. anyway, also, as I understand it, brightness just affects autoexposure. I would probably put brightness at -1 to avoid clipping highlights, but in general i wouldnt use autoexposure much so it wouldnt make much difference. It won't affect latitude.