PDA

View Full Version : Low Light EX1 Overcrank



Elton
11-30-2007, 01:57 AM
Extremely basic shot. My friend Chris let me grab the camera and just chase the kids and see how the slowmo turned out.

Shot at an outdoor mall with only the available light. (dim christmas tree lights mostly) It was 60p for 24 playback and at a 1/120 shutter. No CC--footage was transcoded to ProRes and then h264. Apologies for the fairly heavy compression but file size needed to be kept to a minimum.

Right click, save: http://homepage.mac.com/mrbarlowelton/.cv/mrbarlowelton/Sites/.Public/KidsLoLightSlow.mov-zip.zip

rawfa
11-30-2007, 03:01 AM
Hey Elton, I haven't checked the video yet (downloading) but what do you think of the EX1 so far? I'm very curious about this camera.

Christopher Barry
11-30-2007, 03:06 AM
Thanks for the file, Elton. I can only imagine how good the source file must be before h264 encoding. The EX1 has my attention. What gain setting, if any, in dB?

mikkowilson
11-30-2007, 03:13 AM
Good looking footage there. Thanks for posting. :thumbsup:

- Mikko

Elton
11-30-2007, 06:43 AM
0 dB, I believe. Default camera settings

Elton
11-30-2007, 09:03 AM
Hey Elton, I haven't checked the video yet (downloading) but what do you think of the EX1 so far? I'm very curious about this camera.

Hey rawfa,

The EX1 is a really good performer. It was amazing to be shooting at that shutter speed in low light and still have decent exposure. Good latitude, clean blacks and the resolution is great. XDCAM HQ codec is solid too.

Aerialsfilm M*
11-30-2007, 09:59 AM
Adorable children! :)

Elton
11-30-2007, 12:00 PM
Thanks for the file, Elton. I can only imagine how good the source file must be before h264 encoding.

The original source is great, even when converted to ProRes. It's impressive that it does good 1080p, but I'm actually more impressed with the 720 mode. Very low compression and really clean with 1-60 fps available.

Christopher Barry
11-30-2007, 01:57 PM
Elton,

0dB, wow! And yes, at 1/120, I was surprised when reading the thread initially. No rolling shutter issues, and such a shutter speed would assist in preventing that in this instance.

Any chance of a few uncompressed still frames from the shoot? I would like to see how robust the image is to grade via a Curves filter. Thanks

Christopher Barry
11-30-2007, 02:32 PM
The original source is great, even when converted to ProRes. It's impressive that it does good 1080p, but I'm actually more impressed with the 720 mode. Very low compression and really clean with 1-60 fps available.

Elton, I was reading some of the comments by yourself and others at DVinfo and if I understand correctly, shooting EX1 1080p or 720p, the same data rate of 35megabits VBR in HQ mode is applied, thus you should get cleaner footage from 720p as the EX1 is applying the same compression data rate over less pixels, which should result in less compression. Is that a correct analogy and apply to your statement above?

Barry_Green
11-30-2007, 02:57 PM
and such a shutter speed would assist in preventing that in this instance.
Shutter speed has nothing to do with rolling shutter. A faster frame rate would help assist in preventing a rolling shutter effect (as in, rolling shutter is more prevalent at slower frame rates than at faster frame rates) but the shutter speed is irrelevant. You'll get as much rolling shutter at 1/2000th as you will at 1/24th.

Christopher Barry
11-30-2007, 04:50 PM
noted, thanks.

Elton
11-30-2007, 06:44 PM
Elton, I was reading some of the comments by yourself and others at DVinfo and if I understand correctly, shooting EX1 1080p or 720p, the same data rate of 35megabits VBR in HQ mode is applied, thus you should get cleaner footage from 720p as the EX1 is applying the same compression data rate over less pixels, which should result in less compression. Is that a correct analogy and apply to your statement above?

Yes, that's correct. 720 24p has more than twice the bandwidth per pixel than 1080 24p.

Christopher Barry
11-30-2007, 07:15 PM
This 720p vs 1080p compression analysis should make for some interesting tests in the future, especially low light shoots. Even though the EX1 is proving to have a very sensitive CMOS, getting the best out of the EX1's codec in HQ mode will be an interesting exercise to learn by experimenting/testing.

Stevet
11-30-2007, 08:09 PM
The 1080P stuff looks really good - so much detail, but I must admit the 720P stuff looks magical off this camera, very clean looking.

Christopher Barry
11-30-2007, 08:42 PM
It sure does, Steven. Interested to see how 720p compares to 1080p for shadow detail, and sub 40 IRE range setups. Are you seeing any benefits, Steven?

ullanta
11-30-2007, 09:23 PM
Shutter speed has nothing to do with rolling shutter. A faster frame rate would help assist in preventing a rolling shutter effect (as in, rolling shutter is more prevalent at slower frame rates than at faster frame rates) but the shutter speed is irrelevant. You'll get as much rolling shutter at 1/2000th as you will at 1/24th.

This seems counter-intuitive... can you explain a bit more? That is, how exactly the rolling shutter is implemented that would make the effects insensitive to shutter speed?

It seems to me that at a 1/2000 shutter speed, all data is read of the sensor in 1/2000th of a second, and that would result in less motion (and less rolling shutter distortion) than if the data is read off at, say, 1/60th for the same camera/subject motion. What you are saying seems to imply that the "shutter" sweeps down the image at a fixed rate for a given frame rate, with each pixel (or line?) sequentially accumulating light for a time period determined by the shutter speed?

I guess the question is, on a CMOS sensor, how are the start and end of light accumulation triggered for each pixel; how is the data scanned off the chip, and how do the two processes interact?

I'm of course off to research this... but if you have a clear explanation handy, please let us know...

ullanta
11-30-2007, 09:51 PM
Interesting - it seems the electronic rolling shutter is an exact analogue of the mechanical rolling shutter - that is, there are two sweeps through the image; one to reset each pixel and one to read out the accumulated voltage. The time (seemingly, measured in rows at a fixed time per row) determines the shutter speed - just like the gap in a mechanical fixed-speed rolling shutter. That is, the time between when the reset sweep and when the read sweep are started determine the exposure time, while the actual sweep speed remains constant.

However, in a variable frame rate camera, it seems that the time to sweep through each frame is certainly variable; one might surmise that it COULD be possible to sweep the image faster at a higher shutter speed. Perhaps this is the kind of thing RED (and perhaps Sony) are doing to minimize the rolling shutter effects...

Barry_Green
11-30-2007, 10:06 PM
This seems counter-intuitive... can you explain a bit more? That is, how exactly the rolling shutter is implemented that would make the effects insensitive to shutter speed?
The rate that the rolling shutter "rolls" down the frame is set by the frame rate, not the shutter speed. The rate that the rolling shutter exposes each line is set by the shutter speed. The shutter will sweep the frame in 1/24th of a second, or 1/30th, or 1/60th depending on your frame rate and regardless of shutter speed.


What you are saying seems to imply that the "shutter" sweeps down the image at a fixed rate for a given frame rate, with each pixel (or line?) sequentially accumulating light for a time period determined by the shutter speed?
That's my understanding of it, yes.

Russ Andersson talks about it more here:
http://cybermessageboard.fatcow.com/ssonte/viewtopic.php?t=478&postdays=0&postorder=asc&start=15&sid=49a7b8cfcd77cb7dcd735c4b175c1802

Barry_Green
11-30-2007, 10:11 PM
Interesting - it seems the electronic rolling shutter is an exact analogue of the mechanical rolling shutter - that is, there are two sweeps through the image; one to reset each pixel and one to read out the accumulated voltage. The time (seemingly, measured in rows at a fixed time per row) determines the shutter speed - just like the gap in a mechanical fixed-speed rolling shutter. That is, the time between when the reset sweep and when the read sweep are started determine the exposure time, while the actual sweep speed remains constant.
Yes, but with a MASSIVE difference -- the mechanical shutter exposes the entire frame for a large part of the exposure time. The CMOS rolling shutter doesn't, it only exposes a line at a time. Huge difference. A film camera will have global exposure for almost the entire shutter time, and thus will have the signature motion blur we come to expect from a film camera. And the entire frame is exposed all at once for most of the exposure time. Even the top and bottom of the frame will be exposed simultaneously on a rolling mechanical shutter for some amount of time.

A CMOS rolling shutter won't have that, it'll have different exposure completely. The top will be started and finished long before the bottom even starts. Totally different process, and those who say a rolling CMOS shutter is "just like film" either have no idea what they're talking about, or ... well, can't really think of an "or" to that! :) It's possible to get a little bit of leaning out of a film camera if you pan very quickly, but nothing like the "skool" video would ever happen on a film camera.


However, in a variable frame rate camera, it seems that the time to sweep through each frame is certainly variable; one might surmise that it COULD be possible to sweep the image faster at a higher shutter speed. Perhaps this is the kind of thing RED (and perhaps Sony) are doing to minimize the rolling shutter effects...
I talked to the guys from Thomson about how they were dealing with it, and basically they run the camera at double speed internally. So when set to shoot 60i, internally it's actually running at 120fps and they just discard every other frame. By cranking up the frame rate, that inherently speeds up the "sweep" rate. I don't know how Red is trying to address the issue. I guess it's possible Sony might do something similar, who knows... maybe they employ double-speed sweeping up to 30fps (i.e., internally running at 48fps when recording 24p, and discarding every other frame) but once they hit 30fps or more, dropping to no double-sampling? Or maybe internally the thing's running at massive speeds, 120fps or 240fps, and thus using only one out of every two or four frames, but using that cranked up frame rate to get a faster sweep rate?

If so, though, you'd expect a limit on exposure time based on the frame rate... if they were running 120fps internally to generate 60fps, you should theoretically not be able to set a shutter speed slower than 1/120th. At 120fps the "sweep rate" must be completed in 1/120th of a second, so you couldn't go exposing 60p at 1/60th in that scenario. So, question for EX users: can you set a shutter speed of 1/60th when shooting 720/60p? If so, that would seem to imply that they're not using a doubled (or more) internal frame rate.

Wouldn't it also seem to mean that the rolling shutter lean/wobble would be a lot more noticeable at slower frame rates... If you're running 2 fps at 1/2 second, the sweep would have to be no faster than 1/2 second, yes? Therefore massive leaning should happen, should it not? Of course, much of the lean would be hidden under the blur of a 1/2-second exposure too.

ullanta
12-01-2007, 12:45 AM
Yes, but with a MASSIVE difference -- the mechanical shutter exposes the entire frame for a large part of the exposure time. The CMOS rolling shutter doesn't, it only exposes a line at a time. Huge difference. A film camera will have global exposure for almost the entire shutter time, and thus will have the signature motion blur we come to expect from a film camera. And the entire frame is exposed all at once for most of the exposure time. Even the top and bottom of the frame will be exposed simultaneously on a rolling mechanical shutter for some amount of time.

I certainly don't know the details of what specific cameras do - and what you say certainly seems true of some of the more extreme cases we're seeing. However, the clearest explanation I've seen ( http://www.isgchips.com/pdf/Shutter_Operations_Kodak_App_Note.pdf ) contradicts this statement - the time between the 2 sweeps determines the shutter speed, and the "time" (measured in lines) may be as much as the full frame.

In your scenario, exposure of, say, a 1080p24 image would be (1/24)/1080 - that is, about 1/24000 - at best. So I think it makes sense that with an electronic rolling shutter, too, the much of the frame is exposed simultaneously. For 1/48, half the frame would be exposed at all times, etc.

Having never laid eyes on the shutter mechanism of a film camera, I'm also surprised when you say "the entire frame is exposed all at once for most of the exposure time"... I thought the shutter angle was proportional to the amount of exposure time, with 360 meaning no exposure, and 0 meaning no occlusion - this would indicate that, for non-0 angles, the entire frame would never be exposed at once -and at common settings of around 180 degrees, only about half the frame would ever be exposed at once...?

ullanta
12-01-2007, 12:55 AM
By cranking up the frame rate, that inherently speeds up the "sweep" rate. I don't know how Red is trying to address the issue. I guess it's possible Sony might do something similar, who knows... maybe they employ double-speed sweeping up to 30fps (i.e., internally running at 48fps when recording 24p, and discarding every other frame) but once they hit 30fps or more, dropping to no double-sampling? Or maybe internally the thing's running at massive speeds, 120fps or 240fps, and thus using only one out of every two or four frames, but using that cranked up frame rate to get a faster sweep rate?

If so, though, you'd expect a limit on exposure time based on the frame rate... if they were running 120fps internally to generate 60fps, you should theoretically not be able to set a shutter speed slower than 1/120th. At 120fps the "sweep rate" must be completed in 1/120th of a second, so you couldn't go exposing 60p at 1/60th in that scenario. So, question for EX users: can you set a shutter speed of 1/60th when shooting 720/60p? If so, that would seem to imply that they're not using a doubled (or more) internal frame rate.

Wouldn't it also seem to mean that the rolling shutter lean/wobble would be a lot more noticeable at slower frame rates... If you're running 2 fps at 1/2 second, the sweep would have to be no faster than 1/2 second, yes? Therefore massive leaning should happen, should it not? Of course, much of the lean would be hidden under the blur of a 1/2-second exposure too.

I think they might mean that they are indeed speeding up the sweep rate, and allowing the voltage to accumulate over multiple sweeps - that is, if sweep time for a frame is S, they start the "read" sweep more than S after the reset sweep. This would be equivalent to having the entire frame exposed for some percentage of the exposure time. This isn't exactly "discarding" frames, though, if that's precisely what they meant...

Actually, maybe they are "discarding"... if they are for some reason able to run a reset sweep when they want, but the system requires the "read" sweeps to happen like clockwork once per frame - they can speed up the framerate, but not do the reset every frame. This would allow the voltage to accumulate over multiple read cycles - so, say, at 120 fps, with resets every 1/60th, discarding the result of every other read sweep, you'd have the exact same thing as 60fps, but with a significant amount of full-frame exposure...

blah blah

Christopher Barry
12-01-2007, 02:36 AM
Some interesting info provided by Anders Holck on REDuser forum, discussing how some film cameras using a 45 degree mirror shutter (http://www.reduser.net/forum/showpost.php?p=34308&postcount=42) are effectively exposing film left to right of frame. A visual representation (http://www.reduser.net/forum/showpost.php?p=34857&postcount=53) of mirror shutter as opposed to a rolling shutter.

Back to your thread, Elton. :beer:

Barry_Green
12-01-2007, 09:34 AM
I certainly don't know the details of what specific cameras do - and what you say certainly seems true of some of the more extreme cases we're seeing. However, the clearest explanation I've seen ( http://www.isgchips.com/pdf/Shutter_Operations_Kodak_App_Note.pdf ) contradicts this statement - the time between the 2 sweeps determines the shutter speed, and the "time" (measured in lines) may be as much as the full frame.

Excellent document, thanks for pointing it out, and yes anyone interested in the subject should be reading stuff like that.

I'm basing my statements on observation of what happens on existing rolling shutter cameras. You have to keep in mind that the Kodak document was talking about still camera/DSLR use, not high-frame-rate film use. It's possible to get full-frame exposure depending on how much readout time is available, and DSLRs are slow frame rate, thus they can delay readout until the entire frame has been exposed. In video cameras the readout process is happening simultaneously with the exposure process; lines are being exposed and then shut off and read out while the next set of lines are being exposed.

How much of a "buffer time" can be allocated is probably dependent on price, I would guess. Lower-cost rolling shutter systems are likely to have less buffer time and thus more wobble. Maybe?


Having never laid eyes on the shutter mechanism of a film camera, I'm also surprised when you say "the entire frame is exposed all at once for most of the exposure time"... I thought the shutter angle was proportional to the amount of exposure time, with 360 meaning no exposure, and 0 meaning no occlusion - this would indicate that, for non-0 angles, the entire frame would never be exposed at once -and at common settings of around 180 degrees, only about half the frame would ever be exposed at once...?
360 degrees would mean no occlusion. A 10 degree shutter angle means that 350 degrees is shrouded, with a 10-degree opening. A 350-degree shutter would operate, sweep-style, like a rolling shutter CMOS video cameras that I've explored.

With a typical film camera shutter you're looking at a half-moon shape on a shutter circle that's maybe 2.5 inches in diameter, which unveils a little square of film that's 10mm wide by 7mm tall. That shutter is going to sweep the frame in 1/48th of a second, and unveil the top left corner (which is actually the bottom right, but who cares) first, and sweep diagonally right down the frame at a decreasing angle until the midpoint of the frame at which point it'll be level, and then the angle will reverse as the rotation continues until the whole frame is exposed. And the entire frame will continue to be exposed simultaneously until the gap is past and the opaque surface begins to obscure the frame in the same pattern.

If you're using a tiny gap like 22 degrees, then no probably the entire frame wouldn't be exposed simultaneously. But if you're up around, say, 45 degrees or more, then yes the entire frame will be exposed simultaneously, and at 180 degrees the entire frame is exposed simultaneously for a goodly amount of time.

Elton
12-01-2007, 10:38 AM
To sum up in practical terms my experience with the issue is that it's basically a non-issue---bumping the camera and walking handheld, even running and fast panning doesn't exhibit rolling shutter artifacts anywhere close to the level of the HV20.

Yeah, it can slightly wobble and skew, but it really has to be coaxed out with ridiculously abusive shooting and then micro-analyzing stills from a moving medium.

But, oh yeah, where were we? Ahh...it's damn good in low light.

Stevet
12-01-2007, 12:44 PM
Yes, and under those abrusive beatings, I can't imagine any camera producing lovely images. ;)

Christopher Barry
12-01-2007, 01:52 PM
But, oh yeah, where were we? Ahh...it's damn good in low light.

Shooting 60p and at 1/120 at 0dB, I have to say the outcome of the footage is amazing. I would like to see the footage and compression outcome if the same shoot was at 24p and various shutter speeds, say 1/48, through to 1/120 and then 0db through to 18dB.

The existing footage has a little room to move with grading, however, soon the compression artifacts are revealed if you heavily brighten the image, as I did using a Curves filter. This is not unusual when grading this type of low light shoot. Perhaps shooting 24p and perhaps a combo of slower shutter speed and gaining up, will provide more detail in the darkest areas, and allowing better manipulation in post. Also, a comparison may reveal how robust the new 35mbps VBR codec is. Oh, and then there is all those other settings in the menu, including Cine 1-4 and individual settings, for those with a studio monitor with HD-SDI handy and a DIT approach to setup. Seems like a lot of testing to follow. Thanks Elton.

Tim Le
12-01-2007, 02:19 PM
Shutter speed has nothing to do with rolling shutter. A faster frame rate would help assist in preventing a rolling shutter effect (as in, rolling shutter is more prevalent at slower frame rates than at faster frame rates) but the shutter speed is irrelevant. You'll get as much rolling shutter at 1/2000th as you will at 1/24th.

I found that shutter speed does affect rolling shutter, at least for partial exposure. I ran some tests with an HV10 and a camera flash unit and I found I got more partial exposure at higher shutter speeds than at lower shutter speeds. Very high shutter speeds would only have a bar across the frame showing the flash. At an extremely slow shutter speed, say 1/8th, the entire frame was lit and there was no partial exposure at all. Also, at lower shutter speeds, partial exposure doesn't necessarily happen all the time. I popped the flash 10 times at 1/30 shutter speed and half the time the entire frame was exposed, the other times it only exposed part of the frame. At normal playback speed, you could tell there was a partial exposure, but only briefly. It only occurs over 1, maybe 2 frames. Again, keep in mind this is the HV10 I'm talking about, not the EX1.

Jim Kinsey
12-13-2007, 09:08 PM
My head hurts encompassing 360 degrees. Nice article and thanks for educating us on the rolling shutter.

Jim Kinsey

John Babl
12-15-2007, 06:04 PM
What sort of exposure/shutter settings does the camera offer for time lapse/single frame?

Thanks,
JB

macgregor
12-17-2007, 05:35 AM
Too bad, the link seems spired.

ESTEBEVERDE
12-17-2007, 01:58 PM
Too bad, the link seems spired.


Hey MacGregor.

How's the new project going?

What are you guys shooting on?

macgregor
12-18-2007, 03:45 PM
what project?

ESTEBEVERDE
12-19-2007, 10:01 AM
Sorry, I thought you were making a full length SIMILO?

macgregor
12-19-2007, 10:10 AM
Yes well, weŽll continue shooting as soon as the producer returns. Maybe 2019?

ESTEBEVERDE
12-19-2007, 10:50 AM
Yes well, weŽll continue shooting as soon as the producer returns. Maybe 2019?


:laugh: !!!

What cam are you using this time out?

macgregor
12-20-2007, 06:40 AM
Arri 435

yucateco
01-18-2008, 10:38 AM
link is death :(

Alvise Tedesco
01-29-2008, 01:10 AM
Bring the link back, please.

ps I love Merida, Yucateco