HDR Photography

[purefilm]

I have been a VFX artist for many year and have used HDRI in my 3d package. I have seen the growing use of HDR in photography. I from what i have seen HDR really creates a cinematic look. I have been wondering if it would be possible to create a camera that could shoot The 3 exposure in the camera, instead of manually changing the exposures. If this was possible Could you image the possibilities you could have if you could build a cinema camera with the ability to shoot HDR.

 

[Nathan Buxton]

its possible... but a lot of other features (resolution, framerate) would need to be removed or retarded in order to accomplish that... maybe red three or red four.

 

[Tom Lowe]

You could essentially do this now with RAW, to a certain extent.

I think the problem is, that the software which blends the images using a certain algorithm tends to apply the changes from frame to frame inconsistently, so you get crazy looking results with moving HDR image sequences (video).

 

[David Mullen ASC]

There is always a problem with sequential exposure techniques when recombining them in moving images -- objects move, the camera moves, etc. If people are noticing a minor color fringe with some lens optics or prism blocks, or sawtooth edges when combining interlaced fields, imagine the fringing, smear, etc. when combining sequential images with movement.

Any HDR solution for moving image is going to have to involve simultaneous exposures, whether by having twice as many photosites with 50% of them ND'd, or some other design trick. Maybe someday there will be an 8K RED sensor for 4K output, for example, with half the photosites covered with ND to get more overexposure information. But of course, it is hard enough to get 12 million photosites packed into the area of a 35mm frame and then run them up to 60 fps without problems with heat, etc.

 

[Noah Kadner]

I've wondered about this as well. Let's see just for argument's sake the camera could be programmed to shoot at 48fps, with every other frame alternating exposure by a set number of stops so you had the essential HDR range low to hi. In post you use motion tracking and smoothcam type techniques to combine the alternating frames back into 24 fps- removing the motion artifacts and deriving essentially double the latitude. Of course this is huge huge math but the potential breakthrough in image quality seems worth it.

Noah

 

[Tom Lowe]

There is always a problem with sequential exposure techniques when recombining them in moving images -- objects move, the camera moves, etc. If people are noticing a minor color fringe with some lens optics or prism blocks, or sawtooth edges when combining interlaced fields, imagine the fringing, smear, etc. when combining sequential images with movement.

Any HDR solution for moving image is going to have to involve simultaneous exposures, whether by having twice as many photosites with 50% of them ND'd, or some other design trick. Maybe someday there will be an 8K RED sensor for 4K output, for example, with half the photosites covered with ND to get more overexposure information. But of course, it is hard enough to get 12 million photosites packed into the area of a 35mm frame and then run them up to 60 fps without problems with heat, etc.

Even if you could capture the data, the problem would be that your HDR algorithm combines the 3 exposures slightly differently one from 24p "frame" to the next. As leaves blow and sparkles and reflections and shadows change, the software might apply the "merging" with terrible inconsistency in a 15 second clip, for example.

I've seen an example of an HDR video and it was pure chaos - massive flickering and inconsistencies. In my mind the challege is not capturing the data (because you could bracket 2 or 3 shots just from a single RAW frame now), but the bigger challenge is the software, which would need to have incredible, almost AI-level intelligence to really do this well, it seems.

 

[purefilm]

So what about this, capture the images with red raw, output from redcine with 3 exposures, then we would need some software to compile the images, just like photoshop HDR does for stills. Would that work?

 

[William Robinette]

In my mind the challenge is not capturing the data (because you could bracket 2 or 3 shots just from a single RAW frame now), but the bigger challenge is the software, which would need to have incredible, almost AI-level intelligence to really do this well, it seems.

The software is the problem.

 

[David Mullen ASC]

So what about this, capture the images with red raw, output from redcine with 3 exposures, then we would need some software to compile the images, just like photoshop HDR does for stills. Would that work?

Why would a single RAW capture provide extra exposure information?

 

[Tom Lowe]

Because you can change the exposure with great latitude with DLSR RAW images. In one bracket, the foreground can be exposed properly, in the next bracket of the same RAW image, the mountains will be exposed properly.

 

[purefilm]

It won't be as acuarate as if you would get the images in camera, but I think it could make for some interesting images.

 

[Noah Kadner]

Considering that the RAW file has more latitude than say a DPX does I see no reason why you couldn't sort of fake HDR:

Take for example a shot of a beach at sunset with bright subjects in the foreground. You output a set of dpx's where you have exposed the for maximum details in the sky. Output a second set where you expose the shot for the foreground. Then comp the two halves together into one shot with a traveling matte to combine the two latitudes. Such as:

-Noah

 

[David Mullen ASC]

Sure, there are special cases where you can combine elements from different exposure passes, but the question is really whether you can make it a standard way of shooting to increase dynamic range. Again, I think any system based on sequential image capture is not really practical for motion picture work in general except for fairly motionless shots. You have to design a way of capturing a high and low exposure image at the same time, not one after the next.

For example (forgetting the resolution loss for a moment), imagine someone taking a prism block in a 2/3" HD camera and putting three Bayer-filtered 2/3" sensors on each end of the block, ND-filtered to get three different exposure passes at the same time. That could extend the range captured by several stops.

 

[Poi Boy]

No doubt it can be done..it just doesn't look very good IMO.
Aloha
-A

 

[Jeff Kilgroe]

Considering that the RAW file has more latitude than say a DPX does I see no reason why you couldn't sort of fake HDR:

Yes, REDCODE RAW does have potentially more latitude than DPX (12bit linear vs. 10bit log), but all you would be doing is faking it. True HDR would require multiple exposures captured by the camera, not the same 12bit image output multiple times with different exposure parameters applied to the same base data. A simplistic example would be one shot of an outdoor scene captured with a .6 ND in front of the lens, keeping the image under control, but letting you expose well into the shadows while clipping the highs. Then another identical shot with a 1.2 ND in front of the lens, allowing you to expose into the highlights. Then combine both images for output to 16bit (or greater) TIFF or EXR or a specific HDR friendly format.

And definitely consider what David said. I don't see it being truly practical or even entirely possible. (edit: entirely possible with current RED / REDCODE).

 

[Jeff Kilgroe]

To address the OP, I think it could be possible to create a cinema camera designed to capture HDRI style imagery. Just expanding on what David wrote above.. The most practical implementation would be a conventional 3-chip system, but instead of separate chips or sensor layers for R,G,B channels, you have a high, medium and low exposure sensor. Each of the 3 chips / layers are constructed or tuned to different sensitivity levels. ...Oone with a base rating of ASA 160, another at 320 and the last at 640. Not saying that would be an optimal range, just a simplistic example.

Perhaps that could be a consideration for a future generation of the RED camera. A "3-chip" design that can deliver a range of 24+ stops and rcord to a 32bit version of REDCODE. Ah, it's fun to dream...

 

[Yannick Hagman]

just like photoshop HDR does for stills. Would that work?

Where is this photoshop-filter located?

 

[purefilm]

The fact is that there where people who said the Red was impossible, I don't what to hear the word impossible again. Probable, but not impossible.

Open Photoshop - open your 3 files - file - automate - Merge to HDR

 

[Jeff Kilgroe]

The fact is that there where people who said the Red was impossible, I don't what to hear the word impossible again. Probable, but not impossible.

Sure, it's possible to build a different camera with HDR in mind. But For the current incarnation of RED, it is impossible for it to capture multiple exposure HDR of the exact same instance in time... Unless the Mysterium sensor can warp the space time continuum. Perhaps it can, I haven't seen the full specs. :matrix:

But I see no reason why the current RED couldn't produce HDRI just as a current DSLR would, with subsequent exposures. Does little good for a moving image, but for static imagery the RED could be a powerful tool. Could be quite useful for capturing background plates or other elements for comping into a shot. Then all the aspects of what you shoot match your other RED footage (obviously).

 

[David Mullen ASC]

So far, I have mentioned two possible ways of technically getting simultaneous exposures -- put twice as many photosites in the sensor and ND half of them, or use a prism block and split the light to different sensors, some ND'd for the high end information.

Since we're taking an attitude of "nothing is impossible" maybe someday someone will manage to make Foveon sensors work for moving images:
http://en.wikipedia.org/wiki/Foveon_X3_sensor

Since the architecture is similar to color negative emulsion, maybe someday there will be a way to do what color negative does, which is have a fast and slow layer per color.

I think the idea of a single sensor with half the photosites ND'd is probably the most probable solution (the RGB-striped sensor in the Panavision Genesis does something like that) but then you run into either having to double the size of the sensor (make it something like the Phantom 65 and needing medium-format optics) or cut the resolution in half (35mm-sized sensor but limited in the final RGB image to HD or 2K resolution like the Genesis.)