EPIC HDR...

[Michael Brennan]

I see that Canon have announced a large sensor that is very sensative, full colour HD images under the light of a half moon. I suggested a few years ago that Jim's large sensor could have a varient with pixels the size of golf balls for low light work.
The rather narrow minded posters on other forums procaliam that we only need +3 stops of HDR.
Imagine a night scene where the landscape is perfectly exposed lit by moonlight, but we can pull down the otherwise overblown ambient street lighting.

Go for it Jim!


Mike Brennan

 

[michael zaletel]

PRO HDR iPhone App Raises the Bar for Auto-HDR

PRO HDR iPhone App Raises the Bar for Auto-HDR

I hope RED uses this iPhone App as a point of comparison when testing EPIC HDR in a wide variety of shooting situations. It's by far the best auto-HDR I've ever seen. Obviously it has the disadvantage of working with the $30 iPhone camera images but what it does with those source images in terms of auto-HDR is truly amazing.

PRO HDR
http://itunes.apple.com/us/app/pro-hdr/id347104281?mt=8

-michael zaletel

 

[Michael Hastings]

My guess is that the sensor is being read twice, but not one complete sweep after the first. Its more like each row is read twice in the single scan. Think of two waves sweeping down the sensor, one right behind the other.

No matter how it works, I'm sure RED HDR is second to none!

that's what I mean in my post about line after line but Jeff said Jim has said this isn't it. Still seems like the only logical method to get true multiexposure HDR without increased motion artifacts.

 

[Lauri Kettunen]

that's what I mean in my post about line after line but Jeff said Jim has said this isn't it. Still seems like the only logical method to get true multiexposure HDR without increased motion artifacts.

There is another option nobody has not yet suggested; The preamplifier of the sensor. They might be able to take two readings of every pixel with two different preamplifier settings. And if so, then the user setting is about selecting the amplification in some prespecified limits.

 

[Jeff Kilgroe]

that's what I mean in my post about line after line but Jeff said Jim has said this isn't it. Still seems like the only logical method to get true multiexposure HDR without increased motion artifacts.

Jim has been real vague, and of course they don't want to tell us (or competitors) how they're doing it anyway. But it's been said that there is no temporal shift between any captured exposure information. If that is indeed the case, then reading a line and then reading it again at a subsequent time will not work. I'm pretty sure they have a way of gathering two exposure readings simultaneously.

The preamplifier that Lauri mentions could be a possibility. Also potentially related to that, and possibly another clue, is that we were told a while back that Epic has two native sensitivity options and ranges where it would perform best. The M-X sensor likes to work around ISO 800-1000. But it's been mentioned that people would be able to have their Epic changed by RED to like a lower ISO range, like 50-250 better. Perhaps that ability has been incorporated to some extent and is able to operate simultaneously in some way, at the penalty of double the data being captured.

 

[Lauri Kettunen]

Also potentially related to that, and possibly another clue, is that we were told a while back that Epic has two native sensitivity options and ranges where it would perform best. The M-X sensor likes to work around ISO 800-1000. But it's been mentioned that people would be able to have their Epic changed by RED to like a lower ISO range, like 50-250 better.

Jeff, that is my point. In my understanding changing the ISO is typically about changing the amplification in CMOS cameras except R1. It's just called ISO due to historical reasons. In addition, if I've got it right, the sensitivity of the pixels cannot be changed. However, after pixels there's all sort of things in the sensorchain including the preamplifier, and the optimal amplification level need not to be unique. Now, you say there are two optimums, and perhaps they just make a very good use of that.

 

[Lee Fingersh]

Jeff, that is my point. In my understanding changing the ISO is typically about changing the amplification in CMOS cameras except R1. It's just called ISO due to historical reasons. In addition, if I've got it right, the sensitivity of the pixels cannot be changed. However, after pixels there's all sort of things in the sensorchain including the preamplifier, and the optimal amplification level need not to be unique. Now, you say there are two optimums, and perhaps they just make a very good use of that.

Turning down the gain on saturated sensor cells during read out won't recover the count of the photons that struck the sensor while they were saturated.

 

[Hrvoje Simic]

Audio world:

http://en.wikipedia.org/wiki/Dynamic_range_compression

In simple terms, a compressor is an automatic volume control. Using downward compression, loud sounds over a certain threshold are reduced in level while quiet sounds remain untreated. Upward compression involves making sounds below the threshold louder while the louder passages remain unchanged. Both reduce the dynamic range of an audio signal. This may be done for aesthetic reasons or to deal with technical limitations of audio equipment, which is seldom able to cope with the dynamic range the human ear can tolerate.

 

[Steve Wake]

Jim has been real vague, and of course they don't want to tell us (or competitors) how they're doing it anyway. But it's been said that there is no temporal shift between any captured exposure information.

That was my understanding too, then in Jim's post that opened this thread he said:

We have successfully tested EPIC HDRx™ to 18 stops of dynamic range in motion. You can use the mode in two ways... Easy HDR (in camera) and HDRx™ (manipulate in post). It can be used several ways, including the "magic" way... which is very cool. Or there is a "normal" way to deal with motion blur.

[My emphasis.]

So this could be parsed several ways, but the way I read it, there is an issue with motion blur. If there is no temporal shift, why would motion blur be an issue? And why would there be a "magic" and a "normal" way to deal with it?

What I remember Jim saying was there were not two sequential frames with two different shutter speeds. I don't have enough imagination to figure out other ways of getting a temporal shift, but maybe others can try.

In any case, I'm sure it will work worlds better than that iPhone app. If you're wondering why a lot of us are excited, download that sample photo of the beach. Look at the histogram! Then slap a curve layer on there in Photoshop, tighten up the black and white points, bend the curve into an "S" and watch it come alive, without blocking out the shadows or blowing out the clouds. And this is a downrezzed JPEG. Imagine what you could do with raw 3K Scarlet (please) or 5K Epic files.

Oh well, back to work.

 

[Lauri Kettunen]

Turning down the gain on saturated sensor cells during read out won't recover the count of the photons that struck the sensor while they were saturated.

Sure, but that's another issue. Gain is neither the same as the preamplifyer in the sensor. In video cameras usually term gain points to the amplification of the RGB signal. Here the question is of the signal before the AD-conversion. The optimum depends on the inherent noise of the sensor cells and the preamplifyer as this affects how many bits of true information is available.

In short, if the sensor has two optimums, one around ISO 800 and the other around ISO 150-200, and both yield 13 stops of dynamics, in principle their combination should yield more. But, of course, still just guessing and it may be good entertainment to RED to follow all theses guesses and speculations.

 

[Roberto Lequeux]

:rant:

Audio world:

http://en.wikipedia.org/wiki/Dynamic_range_compression

[Off-Topic Rant]This is how they get everything in those invasive commercials to be at the top end... If I could punch every single one of those clients or producers... aghh!!! It is MY living room!
Thank god for DVR... Hopefully people will realize that commercials should be inviting, otherwise we'll just pause and FF>> over their foul @#$%[/OTR]

Anyway, about HDR...
Auto? RLY? No way mang!
Auto's got the cooties!
0.02

 

[Hrvoje Simic]

Well, anything can be misused.

 

[Adrian T.]

What I remember Jim saying was there were not two sequential frames with two different shutter speeds. I don't have enough imagination to figure out other ways of getting a temporal shift, but maybe others can try.

I try: Read without reset for the high level layer, then read with reset for the low level layer.

 

[Stefan Christou]

:rant:



[Off-Topic Rant]This is how they get everything in those invasive commercials to be at the top end... If I could punch every single one of those clients or producers... aghh!!! It is MY living room!
Thank god for DVR... Hopefully people will realize that commercials should be inviting, otherwise we'll just pause and FF>> over their foul @#$%[/OTR]

Anyway, about HDR...
Auto? RLY? No way mang!
Auto's got the cooties!
0.02

It's also how radio sounds good. If you've ever listened to a pirate station and wondered why their levels sound al over the place, it's because they don't have a hardware compressor.

Stefan

 

[Roberto Lequeux]

True, true, just like tone-mapping.

 

[Lee Fingersh]

Sure, but that's another issue. Gain is neither the same as the preamplifyer in the sensor. In video cameras usually term gain points to the amplification of the RGB signal. Here the question is of the signal before the AD-conversion.

Same problem - it's still after the collection of the photons, their conversion to electrons, and the storage of the electrons in the wells. Once those wells are saturated, you lose track of the photon count and not much can be done about it.

You'd have to either collect them more slowly (ND filter - doesn't help with DR because it affects noise in the shadows too), collect them in a non-linear way (film does this), or read out and reset the cells before they become saturated.

 

[Wolfgang Woehl]

or read out and reset the cells before they become saturated.

Bingo-ish. Google cmos and read-out strategies and follow the tracks, lots of (rather technical) papers available. CMOS has evolved quite a bit. Probably RED got in on something promising early so they will deliver now what others undoubtedly will deliver in 2011 or 2012.

Also: engineering, not magic pixel dust.

Btw, I think this is a perfect example of how traditional nomenclature (like shutter) -- while helping people bridge technological landslides -- occludes essential properties of new technologies. Once we stop thinking shutter, or ISO for that matter, we can get a grip on this.

 

[Jan Reiff]

http://www.sovietmontage.com/2010/09/09/hdr-video-a-reality/

 

[Jeff Kilgroe]

http://www.sovietmontage.com/2010/09/09/hdr-video-a-reality/

Oh no... Not that terrible, horrible, dreadfully tone-mapped composited mess of DSLR moire and noise again...


Anyway, not going to get into the details of how or start speculating along those lines. But I still think that RED is able to get two separate readings for each pixel on the sensor in one reset cycle. They may do a read pass and then re-read, before an actual reset, or something to that effect, but for all intents and purposes, it's 2 separate exposures per sensor reset. Motion blur will not be an issue except possibly under extreme motion circumstances and possibly not even then. Jim did make a comment about motion blur a ways back somewhere and it was quoted again a few posts back. That seems to go against some of what was said before about no temporal difference....

Anyway, can't wait to play with this stuff some day. But I'll gladly take my Epic-X's now and wait for HDR to come with an update. ...Just sayin'

 

[Josh C]

Btw, I think this is a perfect example of how traditional nomenclature (like shutter) -- while helping people bridge technological landslides -- occludes essential properties of new technologies. Once we stop thinking shutter, or ISO for that matter, we can get a grip on this.

Totally agree. Unforetunately for me, I've never worked with film in stills photography or film because of restrictive cost of processing. I've learnt everything in digital and it often takes me ages to get my head around something because of this "hangover" terminology. That's both with cameras and also editing software which seem to keep terms from when people were sticky taping things together (and therefore make no sense when using a mouse and keyboard).