Raw Footage

[Chris Forbes]

Does anyone know how far you can "push" REDCODE. Meaning if you had a scene that needed another stop or two could you get away get away with it in REDCINE? Without degradation of the image.

I ask because DSLR's shooting raw can usually get away with 2 stops.

[Evin Grant]

All the posted frames and clips were originated in RAW. Some highlight recovery is possible but current S/N ratios are based on RAW.

[Brook Willard]

I couldn't resist going back to the old charts to find out for myself. Note that these stills were from the old "8 stop" days, not the "11.3 stop" present.

This is both a crop and significant scale from the original 4K files. "Native" exposure for these charts would be about a F:11/16 split.

[Stephen Williams]

I couldn't resist going back to the old charts to find out for myself. Note that these stills were from the old "8 stop" days, not the "11.3 stop" present.

This is both a crop and significant scale from the original 4K files. "Native" exposure for these charts would be about a F:11/16 split.

Hi Brook,

Hopefgully charts showing 11.3 will be shown at NAB.

Stephen

[Chris Forbes]

The reason I ask is because when I shoot RAW with a DSLR I always process the file twice once exposing to bring out detail in the blacks (blowing out the highlights) and once to bring out detail in the whites (crushing the blacks) then composite the two together in Photoshop. Because I can get up to 2.5 stops in either direction without any loss of quality to the image it gives me a much higher range. (I hope I explained that right)

If I could do that with RAW from the RED I could get up to 15 stops of latitude instead of 11. Process the same file twice and luma mask them together.

[Zach Hilton]

Interesting concept. You'll have to keep us posted on how that pans out.

[Brook Willard]

The process you are describing can be achieved with a good curve in REDCINE.

[Billy Summers]

oh boy!

[Chris Forbes]

The process you are describing can be achieved with a good curve in REDCINE.

I don't understand. But I guess @ NAB "all will be revealed"

[Gavin Greenwalt]

What he means is: Ripping a photo in half and recombining it is useless. The data that exists, exists. You aren't doing anything by rendering out two 'exposures' from the same data. Just apply whatever it is you want to do to the footage directly in the form of a curve. This is true if your still photos too. I'm not sure what you think you're accomplishing by exporting two raw exposures.

It's exactly the same as taking a photo, copying it, applying a +10 brightness to one and -10 brightness to the other and blending them back together.

[jbeale]

If you are comparing the JPEG generated by a DSLR to the RAW file that you've converted twice, yes there can be a difference, but it's because of limitations of the software (or in-camera firmware) generating the JPEG. Now REDCINE (from what I gather) is more flexible so you can accomplish the same output all in one single step. (With suitable software you could do the same with your DSLR stills, by the way.)

[Poi Boy]

You can indeed extend the range of a still by combining two raw exposures but it always looks a little funky. I much prefer a good curve; I think the same will be true of Red footage.
Aloha
-A

[Chris Forbes]

What he means is: Ripping a photo in half and recombining it is useless. The data that exists, exists. You aren't doing anything by rendering out two 'exposures' from the same data. Just apply whatever it is you want to do to the footage directly in the form of a curve. This is true if your still photos too. I'm not sure what you think you're accomplishing by exporting two raw exposures.

It's exactly the same as taking a photo, copying it, applying a +10 brightness to one and -10 brightness to the other and blending them back together.

Not quite because of the way a RAW file saves itself. Amazingly enough this thread

http://www.reduser.net/forum/showthread.php?t=630

is similar to what I'm talking about.

Have you worked with RAW digital photos? The exposure is not fixed until you process it. The dynamic range is fixed but the exposure can be moved up to two stops higher or lower than it was shot. This is not the same as adjusting the exposure after it has been fixed (with levels or curves ) in fact I don't know how it could be more different.

This is a great article on the advantages of shooting in Raw.

http://www.rangefindermag.com/Magazine/April04/raw.tml

[Brook Willard]

You must remember that we are not finishing in 8-bit [as most photographs do]. Many [if not most] people will finish in 12-bit.

If you are conforming to 8-bit for any reason, you're only a LUT away...

[Chris Forbes]

You must remember that we are not finishing in 8-bit [as most photographs do]. Many [if not most] people will finish in 12-bit.

If you are conforming to 8-bit for any reason, you're only a LUT away...

Exactly my point.

[Brook Willard]

But of course. It was a more general post.

[Floris Liesker]

Not quite because of the way a RAW file saves itself.
http://www.rangefindermag.com/Magazine/April04/raw.tml

I have to agree with Imthatoneguy in this case, Chris. Data = data, and there is nothing special about the RAW data that makes it possible to push or pull a few stops. It's just that when it's been converted to 8-bit JPG or TIFF you will not have those options without running into quantization artifacts. So it's not a matter of RAW being so special, it's a matter of JPG and TIFF being not special enough. And as Brook pointed out, in 12-bit you can do what you just described easily with a good curve.

[Poi Boy]

I'm not a fan of combining multiple raw files to extend range because I don't like the look, but it does work. Ideally you've captured different exposures to begin with but if you have not, you can push and pull from one raw file and extend the range. If you don't believe it you just haven't tried it or read about it so quit saying it doesn't work... it just doesn't look great.
Aloha
-A

[Chris Forbes]

thank you Poi boy. First let me say thank you for everyone that replied. I think we are debating apples and oranges here (Pictures and Video) so there is room for discussion, but I don't think any of us will know until we have footage in hand to play with.
Second let me say I was a Photoshop lackey for an all digital portrait studio for 6 years. All day every day working with images on the computer. So I have a little experience with RAW image files. I am not an expert in all things but I do know which end of the pen goes on the wacom pad.

So I can tell you that you can extend the dynamic range of an 8 bit .tiff file with the process I described. (and if you don't overdo it, it looks good)

I do not know if that will work with 12 bit RAW video. I have never worked with RAW video because it is brand new but I am looking forward to finding out.

[Poi Boy]

Thanks Chris, I totally agree. I was just trying to make the point to those saying that it doesn't work that it does work.
Aloha
-A

[Chris Forbes]

Thanks Chris, I totally agree. I was just trying to make the point to those saying that it doesn't work that it does work.
Aloha
-A

I know that second part wasn't directed at you.

[Floris Liesker]

Chris, the method you describe sure works, I am just saying you could do exactly that with a curve. And that it is faster too. I'm willing to pick up my wacom-stick and battle it out: Give me an original RAW-file and an end-result. Describe what you have done with it to me in exact terms, especially how you tweaked the luma-key and I'll return to you a curve that will do exactly that. Note: This only goes for the process you described of pushing shadows, bringing down highlights and luma-keying them over eachother. I'm not talking spatial mediation in HDR processing and stuff like that.

My point is: you're not getting anything more out of your footage than you would using a curve, so 11 stops of lattitude will not be 15 all at once, like you suggested earlier in this thread.

[Andrew M.]

Is REDRAW derived from Misterium already after Gamma curve is applied?
In another words, is Misterium converting linear voltages from the sensors (pixels) to the linear capture right in the on board hrdware?
See page 2 bottom in this article.
http://www.adobe.com/digitalimag/pdfs/linear_gamma.pdf

Andrew

[Nick Shaw]

As I understand it, REDCODE RAW is 12 bit linear, therefore totally unprocessed from the output of the 12 bit A/Ds. Likewise the RAW output from the high speed port.

[Andrew M.]

So as per above article half of the 4,096 brightness levels are wasted to represent 20% of highlights and only 64 levels are used to represent 20% of the shadows. Would be much more efficient to apply the gamma curves right on the sensor hardware and then use A/D converter only after gamma curve is applied?

[GlennChan]

So as per above article half of the 4,096 brightness levels are wasted to represent 20% of highlights and only 64 levels are used to represent 20% of the shadows. Would be much more efficient to apply the gamma curves right on the sensor hardware and then use A/D converter only after gamma curve is applied?

Gamma correction (if any) is typically applied right after the A/D converter, then it feeds into the DSP, and into compression and storage.

With Red, no digital signal processing is applied since it introduces noise into the signal and can clip information.

2- The traditional thinking is that applying gamma correction (or things similar to it) makes for more efficient compression + storage. This was originally Red's plan (Redraw to 10-bit log). However, this was changed to 12-bit linear since (according to Graeme) it provides better performance. I don't believe he gave any details on why, but I guess all you need to know is that it works better!

*12-bit linear = no gamma curves applied.

2b- In particular, see this thread on dvinfo:
http://www.dvinfo.net/conf/showthread.php?t=79126

David Taylor, who works on Cineform, is surprised (like I was) that Red was going with 12-bit linear.

3- Recall that the footage is going to be compressed. So by throwing more data in (instead of reducing it by going to 10-bit log), you need 'heavier' compression. So it isn't like 12-bit linear is clearly better than 10-bit log because 12 is a bigger number than 10. You have trade offs.

[Andrew M.]

So it looks like we have only 64 levels of resolution in first 20% of light intensity detection range. Is this enough to show shadows?
I think most still camera CMOS has it inverted, first 20% of shadows is using 2048 levels and the last 20% of highlights is using only 64 levels.
Graeme was explaining this in another thread but not in this context.
Maybe we wait for him to comment...

Andrew

[Andrew M.]

" Last time I checked 8bit log was better than 12bit linear." Nope. 256 levels, no matter how they're spaced, cannot compete with 4096 levels. That's especially the case when the sensor is a 12bit linear sensor.

Gamma encoded 8-bit is fine for delivery for viewing on today's (not tomorrow's) display devices, but it's no good for capture!

-----------------------------
It is what Graeme answered on DVinfo forum back in November about this.
http://www.dvinfo.net/conf/showthread.php?t=79126&page=3
I guess Graeme knows something that we don’t know about linear acquisition from Misterium CMOS.
Why they don’t make CMOS sensors so they are applying gamma curve right on the analog side, can’t get it. It gives minimum 3 extra bits, or as film makers say 3 stops of luminance levels resolution.

I will argue that 9 bit can compete with 12 if it is not done based on reverse luminance levels.

[J. Bernard Vallon]

So it looks like we have only 64 levels of resolution in first 20% of light intensity detection range. Is this enough to show shadows?
I think most still camera CMOS has it inverted, first 20% of shadows is using 2048 levels and the last 20% of highlights is using only 64 levels.
Graeme was explaining this in another thread but not in this context.
Maybe we wait for him to comment...

Andrew

My understanding is a linear signal is a straight off the sensor electron for electron reading:

Meaning if you add an electron to a photocell, it adds a level, 2453 goes to 2454, etc. Light is logrithmic, meaning if you add a stop of detection you need to be able to detect TWICE the number of electrons, so going from 2048 levels to 4096 means you can detect an additional stop, and that entire extra 2048 levels are devoted to one stop only. In this way you can't 'invert' it and devote more levels to the shadows, linear is linear.

DSLRs have no trouble detecting detail in the shadows, for the most part, and because they originate on a linear signal, they have only 64 levels devoted to the shadows as well. Some cameras have 128, some have less. It works fine for the most part, and the mysterium supposedly has real low noise, which means more of that shadow information is usable.

A question i have always had is: if the sensor has a 'native iso', than is there any real benefit in changing to a different iso if your shooting RAW? In fact, wouldnt you only get the full benefit of RAW if your shooting at the absolute lowest ISO available?

[Graeme Nattress]

All I can say is that it's terribly complex :-) And along with that, there's how the RAW data gets converted to RGB, and that must occur with linear data, and theres how the compression works, and that the sensor is inherently linear (as pretty much most sensors are).

In the shadows, it all comes down to SNR, and if you have enough bits so that quantisation noise is lower than signal noise, then yes, you have enough levels to represents shadows accurately.

[Gavin Greenwalt]

Chris, there is nothing magical about your RAW data. You will get no additional data from rendering two seperate files out of a single RAW file and recombining. You will, however, lose data. This is true of still and motion files, since all you usually work with are a sequence of DPX files.

12 Bit RAW -> 12 Bit RGB. All you're doing with the Raw file when you output to RGB and adjust the exposure is adjust the offset of the data. When you do this, if you've maximized your 12bit RAW file, you're clipping data. If you haven't maximized you're just moving it around on your histogram.

What you're suggesting is like exporting two files from an uncompressed source. One in JPG and one in PNG and then recombining them later and suggesting it's more accurate than the uncompressed. I agree it will look different. It will lose data in the process.

Another way to look at it is like taking an HDR image. Exporting two 8 bit JPGs and recombining the JPGs into an HDR image. Why Export and recombine!? You already HAD an HDR image!

[Andrew M.]

In the shadows, it all comes down to SNR, and if you have enough bits so that quantisation noise is lower than signal noise, then yes, you have enough levels to represents shadows accurately.

I remember you were explaining it in another thread.
If noise coming out of the CMOS sensor is 1/4096 of the maximum signal what is the point of going lower then 1/4096 with level measuring.

So going log 12 bits as oppose to linear, I suppose, will just add to the overhead and cost.
Also, converting log to RGB has more overhead then linear I think.
So the logic is, get the data out of CMOS as fast as you can with as little on the CMOS conversion as possible. Once we are out of the camera in the fast computer we have plenty of CPU power to do data massaging.
Yes, we need fast interface on the camera to get all this 12 bits and 4K and 60p but this is less then 10Gbits/sec so it is doable.
Correct?

[Graeme Nattress]

Andrew, that sounds pretty close. We can get the 12bit data out, but compressed via REDCODE RAW quite easily up to 30p. To go beyond that to 60p or up to 4.5k, we need the high speed port.

Graeme

[Mark L. Pederson]

Graeme -

What is the file format of what is coming out the high speed port?

And does this HAVE to go through REDCINE for some type of transcode to be used by 3rd party apps?

[Graeme Nattress]

Well, you could always just write your own RAW decoder as it's just RAW data. File format is still being worked on. REDCINE would be easiest though.

Graeme

[Mark L. Pederson]

cool -