More HDRx...

[David Quinn Carder]

HDR tone-mapped image looks much better than the image @ ISO 320 (rated M sensor exposure index).

As usual I strongly disagree with your aesthetic judgement. That aside… RED posts an amazing demo of HDRx, and you post… mediocre demonstrations of tone-mapping? What's your point?

 

[Sanjin Jukic]

As usual I strongly disagree with your aesthetic judgement. That aside… RED posts an amazing demo of HDRx™™™, and you post… mediocre demonstrations of tone-mapping? What's your point?

David,

if you think it's "mediocre" then you should post here something of your "genius" shots...

Just waiting... :laugh:

 

[Shane Betts]

People have been shooting anything that moves since the inception of the photo-chemical process... So what. The very first moving images ever shot were of common things and they were sought after the world around.

Why be intimidated by tech that makes imaging easier? I don't get the point. To me, it is so trivial that I don't understand the need to even discuss it.

What is remarkable is that not only will we be getting digital tools that really rival film, but we will also have the ability to to manipulate our moving images in much the same way as we do stills in Photoshop... Without a million dollar investment.

Hey Scott, I think we're saying the same thing :cheers2:

A lot of professionals get jittery at the first signs of this kind of technology and start waiting for the industry sky to fall. "Now anybody will be able to do my job". I'm just pointing out that, yes there will be a bunch of folk shooting and getting 80-90% images with this technology but that it takes talent and experience to get those last few percentage points on the screen. I'm also pointing out that, without the need to spend an hour lighting just to get an exposure, we can now concentrate on using lights purely for shaping the image - purely for that last few percentage - and I think that's cool!

I'm not intimidated, I'm as excited as hell!

 

[David Quinn Carder]

David,

if you think it's "mediocre" then you should post here something better of your own shots...

Waiting... :laugh:

I don't have a RED camera and I've never tone-mapped anything. If I did, I still wouldn't post it here because I don't see the point. That was my question: What's your point? What are you trying to demonstrate? Everyone here has seen tone-mapped images, and everyone in this thread is here to see and discuss HDRx.

 

[Sanjin Jukic]

The point is to try out to work on HDR images that can be generated from existing M and MX RED cameras,

get feeling about that and slowly prepare yourself for Epic/Scarlet HDRx.

As Jim already mentioned it could require also some of fine tuning in post.

For that kind of tunes you can learn even before you put hands on new RED HDRx hardware/software...

With this sort of experiments you can also improve a knowledge more about good and bad sides of MX sensor...

It's not a provocation, it's a real and serious thing.

More or less that's it.

 

[Jeff Kilgroe]

The point is to try out to work on HDR images that can be generated from existing M and MX RED cameras,

Sanjin, I know you're trying to help. But what you are doing probably does belong in a different discussion thread. And the only way to actually generate HDR from existing cameras is to place two of them in a mirror rig or to capture subsequent exposures with different exposure time, or amount of ND filtering. What you are demonstrating here is not HDR, it's tone-mapping done by creating two differently processed images from the same R3D data. You are not gaining HDR or anything that is not there in the R3D. Any good RAW processing application can pull up your shadows and bring down your highlights to compress or map your tonal range, without generating two separate false exposure images that you are combining.

This discussion is about HDRx. HDRx is not two differently exposed frames as you are producing. It is a normal R3D stream, properly exposed to maintain the desired level of detail in shadow areas. HDRx provides highlight protection and greater latitude by way of multi-sampling the sensor at several intervals throughout the exposure period, not just on the final read-out and reset action.

It seems this additional data is stored in an "X track", which is not really a secondary image track containing a properly exposed section of bright imagery. Although, such an image could be generated from it. It seems it's something else, not only is it highlight data, but also vector information of sorts for creating the MagicMotion and applying various levels of MagicMotion or MNMB (normal) motion blur.

We don't have all the specifics on how HDRx works, but one thing seems clear. It's not one fast exposure followed by a slower one -- or one exposed frame biased for bright regions of the image and another exposure biased for dragger regions. There seems to be a single "shutter action" in that this all happens within one read/reset or shutter cycle of the sensor. It seems that dialing in your HDRx level, or number of stops of "protection" you want to have, you are in effect dialing in the number of intermediate samples taken during each exposed frame, also allowing for increased precision for the application of MagicMotion.

 

[David Rasberry]

Hey Scott, I think we're saying the same thing :cheers2:

A lot of professionals get jittery at the first signs of this kind of technology and start waiting for the industry sky to fall. "Now anybody will be able to do my job". I'm just pointing out that, yes there will be a bunch of folk shooting and getting 80-90% images with this technology but that it takes talent and experience to get those last few percentage points on the screen. I'm also pointing out that, without the need to spend an hour lighting just to get an exposure, we can now concentrate on using lights purely for shaping the image - purely for that last few percentage - and I think that's cool!

I'm not intimidated, I'm as excited as hell!

Yep, regardless of the tools available, truly talented AND skilled artists are still a rare breed.

 

[Jannard]

After watching "The Social Network" again... apparently there is plenty already to make a great looking feature with a RED ONE MX. Everything we are improving and tools we are adding only gives that much more capability to the program. EPIC is all RED. It is just more and less of everything. More resolution, more dynamic range, less noise, more data if you choose, more frame rate options... all in a smaller package.

Jim

 

[Michael Olsen]

More resolution, more dynamic range, less noise, more data if you choose, more frame rate options... all in a smaller package.

Someone's been listening...

EPIC is shaping up to be a walloping great camera.

 

[Gavin Greenwalt]

It seems this additional data is stored in an "X track", which is not really a secondary image track containing a properly exposed section of bright imagery. Although, such an image could be generated from it. It seems it's something else, not only is it highlight data, but also vector information of sorts for creating the MagicMotion and applying various levels of MagicMotion or MNMB (normal) motion blur.

I doubt the xTrack actually contains a vector pass. Not sure where that data would be coming from. I would assume the xTrack is just a short exposure. The foundry can extract vectors from that, and by being a normal image you get more flexibility in the future when algorithms improve. Baking in vectors would lock you in, which seems pretty un-red-like.

The point is to try out to work on HDR images that can be generated from existing M and MX RED cameras,

But yes Sanjin you are wasting your time by splitting up your image into two and then recombining it. Just apply local tone mapping. That's different from curves that you tried to use as an example of being unable to reproduce "HDR". Curves or LUTs are global tone mapping. Local tone-mapping adjusts contrast based on spatial filters in addition to luminance and chroma. An example of a local tonemapping algorithm would be the "Shadows/Highlights" tool in Photoshop.

Behold the power of tonemapping on a single exposure from REDCine. Same nice and overprocessed look, half the work.:

 

[Jeff Kilgroe]

EPIC is all RED. It is just more and less of everything. More resolution, more dynamic range, less noise, more data if you choose, more frame rate options... all in a smaller package.

Can't wait to get my hands on it!

I doubt the xTrack actually contains a vector pass. Not sure where that data would be coming from. I would assume the xTrack is just a short exposure.

Yes, specific vector information is probably not in the file, but can be extrapolated from multi-frame motion, or possibly within individual frames themselves -- comparing X track sample(s) with the base track or with multiple X track entries. . I'm positive there's more going on than a shorter exposure before the primary exposure..

All speculation anyway... And I'm not expecting anyone from RED to spill the beans anytime soon.

 

[Lee Fingersh]

But yes Sanjin you are wasting your time by splitting up your image into two and then recombining it. Just apply local tone mapping.

There's no possible way that will do the same thing. You have to expose to hold the highlights in the original raw data, whereas the HDR approach will allow you to expose for the shadows and still hold the highlights. That means you'll have 3-6 stops more noise in the shadows at the same light levels using the single-exposure approach. Now, if that's acceptable, that means your scene DR isn't too high to get what you want out of the base sensor performance and that's just fine. But there could easily be times when 3-6 stops more shadow noise will be very unacceptable and you'll wish you could have exposed for the shadows and held the highlights at the same time, which is what the HDR approach does.

 

[Eric Haase]

I doubt the xTrack actually contains a vector pass. Not sure where that data would be coming from. I would assume the xTrack is just a short exposure. The foundry can extract vectors from that, and by being a normal image you get more flexibility in the future when algorithms improve. Baking in vectors would lock you in, which seems pretty un-red-like.

But yes Sanjin you are wasting your time by splitting up your image into two and then recombining it. Just apply local tone mapping. That's different from curves that you tried to use as an example of being unable to reproduce "HDR". Curves or LUTs are global tone mapping. Local tone-mapping adjusts contrast based on spatial filters in addition to luminance and chroma. An example of a local tonemapping algorithm would be the "Shadows/Highlights" tool in Photoshop.

Behold the power of tonemapping on a single exposure from REDCine. Same nice and overprocessed look, half the work.:

Did you do this in Redcine? Can you describe what you did?

 

[Roberto Lequeux]

Gavin, did you shoot that as well? Second time I say this to you Gavin: Good eye! It may not be natural enough for critical eyes, but I think it is beautiful and could fit in certain situations. I am thinking romance. And in all genres not everything has to be natural looking. Pull it back a bit and you could win them all.

 

[Gavin Greenwalt]

Gavin, did you shoot that as well?

No, that's Sanjin's earlier posted/processed R3D. He deserves all credit for the shot.

I just reprocessed it in Photoshop to demonstrate that you don't have to split export an r3d to apply local tone mapping.

There's no possible way that will do the same thing. You have to expose to hold the...

Sanjin was saying you have to process an r3d under two ISOs to get a tonemapped look. The point was to demonstrate that a single export with maximum DnR can still be processed just as well.

Did you do this in Redcine? Can you describe what you did?

No just exported from REDCine flat.
Photoshop: Shadows/Highlights, curves, hue/saturation.

 

[Roberto Lequeux]

That's just Sanjin's earlier posted/processed R3D.

I just reprocessed it in Photoshop to demonstrate that you don't have to split export an r3d to apply local tone mapping.

Gotcha, hence the down-scaling.

Nice frame SJ, nice grade Gavin.

Shadows/Highlights is one of my favorites.

 

[Karel Bata]

Wow this is indeed incredible. Congratulations.

Though I must say I'm not completely surprised. A couple of years back I was experimenting with extracting HDR images from moving image film negatives, with promising results. But then the spec of the Alexa was announced and I realised I was just wasting my time. This development only confirms that!

A few things have struck me here. Like, the choice of subject matter illustrates effectively how well the the new cameras will capture details in the shadows and highlights simultaneously, but that choice doesn't show how well it performs with a full tonal range - in other words the mid-tones are missing and we can't tell from this how well the tone mapping algorithms will cope with a more typical high contrast subject.

The original image here would have had detail predominantly in highlights and shadows and would be distributed something like this (though spread across two files)

tone mapping (I assume) has produced this

. .

and all that has really happened in this case is that the information has been shunted along from both ends towards the middle. So an image that would have displayed on a conventional monitor like this

looks like this

Very nice. But I do wonder what processing an image that had subject matter within the whole tonal range would look like. If there had been some image information in the mid tones how well would that process have worked?

Edit: THIS SECTION DELETED BY POPULAR DEMAND

 

[Peter Majtan]

Karel, You are analyzing compressed JPEG file... And assuming...
All You are doing here is the proving the point of what he is trying to explain in the new thread...

Just commenting... that is one of the pitfalls, someone reads something into everything.

Jim

:sifone: Peter

 

[Graeme Nattress]

You're right Peter - analyzing a compressed JPEG tells you as much about JPEG compression as it does about the image

From what I'm seeing the HDRx files have more DR than I'm used to dealing with, and that's a good thing.

Graeme

 

[Karel Bata]

OK. Forget the latter half. That shows artifacts I've not (knowingly) encountered before. Odd though. I've edited that section out now.

But I still think the subject matter chosen lacks mid-tones and it would be interesting to see what a picture with a full tonal range would yield.

"the HDRx™™™™™ files have more DR than I'm used to dealing with, and that's a good thing." Well yes, I agree entirely.