HDR macro

[Cory Mitchell]

I was looking up the process to capture an hdr image, where you take multiple (5-8?) photos of a mirror ball with different exposures (under exposed all the way to over exposed)

I was wondering if it would be possible to make a small program/macro that would automatically (at 24-60 fps) do this for you with the Red camera.

So basically you focus the RED at a mirror ball hit the button and in a second you have all the pictures you need to make an hdri for the scene. (instead of taking one shot, manually changing the exposure and then taking another shot, etc...

Would stuff like this be possible with the RED?

thanks

 

[Joe Aurili]

That is something the SDK or API should be able to be programmed to perform.

 

[I Bloom]

I was looking up the process to capture an hdr image, where you take multiple (5-8?) photos of a mirror ball with different exposures (under exposed all the way to over exposed)

I was wondering if it would be possible to make a small program/macro that would automatically (at 24-60 fps) do this for you with the Red camera.

So basically you focus the RED at a mirror ball hit the button and in a second you have all the pictures you need to make an hdri for the scene. (instead of taking one shot, manually changing the exposure and then taking another shot, etc...

Would stuff like this be possible with the RED?

thanks

The answer is yes. Definitely possible though at first it might be a bit of a hack.

Most likely the first easy way to do this will be by controlling the camera externally with say a laptop via the USB interface. Eventually my hope is that such programs might be able to reside in the camera software itself, added by enterprising programmers to a camera that is open to such "plugins".

We had a thread about such innovations they would be made possible by a RED SDK (software developement kit):

http://www.reduser.net/forum/showthread.php?t=2597

IBloom

 

[J. Bernard Vallon]

That seems like an unnecessary complicated way to pull HDR source footage from red.

Because red has a fully electronic shutter, you could under/over expose by shutter priority, and shoot at 72fps. 2 stops over, 2 stops under, and correct, alternating. As long as your action wasn't so fast that there was a significant difference between frames you're golden.

f8 1/48th
f8 1/96th
f8 1/192nd
and so on.

 

[Cory Mitchell]

thanks guys for your responses. I'm glad this would be possible with the SDK. As long as someone would actually do it... (I'm not a programmer)

PerfectOptics could you clarify what you mean? I thought the Red could only go 60 fps at 4k (unless you're shooting 720p). more than anything I'm just not sure what you mean. I understand it's an electronic shutter, but how could you automatically get under/over exposed shots without somehow setting it that way? Do mean just changing the frame rates? don't you still have to change the f-stops? Sorry if I'm misunderstanding.

thanks again.

 

[Barend Onneweer]

Assuming you don't want moving images, you can already do this easily by using a DSLR's bracketed exposures function.

Pretty much any DSLR camera has a function that lets you automaticaly snap multiple pictures in bracketed exposures.

Not that it would hurt to have the same functionality on RED One, but I think in most occasions I'd still use my still camera.

Bar3nd

 

[_BK]

HDR can be captured directly by altering the way the CMOS sensor captures its data...basicaly by sub sampling pixels that are over/under exposed. The Phantom guys are currently experimenting with this method on their camera.

There is a white paper out there that explains the process... Ill try to dig it up.

 

[Jim Arthurs]

I still do my HDR enviro's the old school ILM way.... with my Nikon Coolpix 995 and 180 degree full frame fisheye lens... beats the mirror ball approach in every aspect... I just do 3 exposure series rotating on nodal 120 degrees apart.

The resolution this produces is fine for lighting and reflections, but not for actual imagery past SD/720p... which brings us to...

...With RED and 4K, I see a need for VERY large environment maps... much larger than standard DSLR's can record to... not quite sure how to approach this yet...

 

[CVB]

Jim, we were thinking about making a multi-gigapixel imaging script for our moco head. Basically you could input the FOV of the lens you are using and it would automatically map out the surroundings by spinning around to pre-calculated positions and trigger the frames on the camera. The head could go between positions in less than 1 second per move so I'd expect it wouldn't take long to map out an entire environment at very high resolution.

 

[Curran Giddens]

I still do my HDR enviro's the old school ILM way.... with my Nikon Coolpix 995 and 180 degree full frame fisheye lens... beats the mirror ball approach in every aspect... I just do 3 exposure series rotating on nodal 120 degrees apart.

Interesting. I have a Nikon Coolpix 990 and a fisheye lens. I'll have to try this sometime....

Jim, we were thinking about making a multi-gigapixel imaging script for our moco head. Basically you could input the FOV of the lens you are using and it would automatically map out the surroundings by spinning around to pre-calculated positions and trigger the frames on the camera. The head could go between positions in less than 1 second per move so I'd expect it wouldn't take long to map out an entire environment at very high resolution.

Sounds like fun. I'm sure it will be quite useful too.

 

[J. Bernard Vallon]

PerfectOptics could you clarify what you mean?

Sure. To build an HDR image, you need several source frames at different exposures. You could do this by bracketing with your aperture or your shutter speed. (Using your ISO doesn't TECHNICALLY increase your dynamic range, it only gives the illusion of it, because with digital sensors ISO is a post-capture adjustment.)

With a motion picture image, you can't vary your aperture on a per-frame basis, but your shutter is entirely electronic, no physical curtain. The little computer inside your RED will say (60 times per second) "Ok, let the sensor expose for a 1/48th, then record the results". In this case, youd have software installed on the red that alternates exposures in a cycle:

1/50
1/200
1/400
1/50
1/200
1/400
1/50...

Each set of three frames are stitched together to get an HDR frame. You'd have to have a special piece of firmware to pull this off, something like a 3k sample of the sensor at 72fps to get a 24fps resulting image.

Does that make more sense?

 

[_BK]

Each set of three frames are stitched together to get an HDR frame. You'd have to have a special piece of firmware to pull this off, something like a 3k sample of the sensor at 72fps to get a 24fps resulting image.

This method would not produce a very usable image. Even when shooting bracketed exposure HDRIs with a DSLR any movement in frame between shots creates nasty artifacts in the final combined HDRI.. people moving, leaves blowing, ect.

 

[Tom Lowe]

I don't think HDR goes well together with moving images, as a general rule, because the software algorithms mergerging the exposures are going to be all over the map as the images change from frame to frame. I have seen a few examples of video HDR, it was shockingly horrible.

Someone here with more knowledge of recent technology might make me a liar, though.

 

[Cory Mitchell]

Yeah thanks for explaining that PerfectOptics. I was thinking about before and kind of started putting it together.

I guess I didn't clarify. I actually didn't want to make a motion HDR image (although if it worked good that would be cool), just a function to get a quick HDR for any given shot for sfx.

Like you know you're going to have sfx in a shot then you hit this macro button (or whatever) and it takes a bunch of images which you convert to hdr and use later.

The question came when I was thinking about eliminating the need to use a still camera to perform this, and how much easier/quicker it would be if you just hit a button and the RED did it for you in a couple seconds, without needing intervention.

While using a fish eye sounds pretty cool, it would definitely defeat the point, as my original idea was wanting to eliminate having to change stuff around to make the HDRI (changing lenses, etc...).

The context of this idea was basically wanting to get HDRI quickly and without having a big crew.

Either way it sounds like the RED is pretty flexible to do this sort of stuff, which is pretty exciting.

 

[Gavin Greenwalt]

Jim, we were thinking about making a multi-gigapixel imaging script for our moco head. Basically you could input the FOV of the lens you are using and it would automatically map out the surroundings by spinning around to pre-calculated positions and trigger the frames on the camera. The head could go between positions in less than 1 second per move so I'd expect it wouldn't take long to map out an entire environment at very high resolution.

Any chance that might be ready to roll by say... late august?

 

[Cory Mitchell]

There was a program I found a while back that could take a video where you pan and then it would convert that to a panoramic image.
It was really cool, but it was buggy, and then seemed to fall of the face of the earth. It also only could handle DV resolution, but it would be ultra cool to use with 4k or even HD (1080)

 

[Jeff Brue]

Really guys take a look at the Vision Research white paper it has an excellent feature that slides the recording bracket for a photosite based on the exposure of the pixel ie its designed to record and gather valuable data for flaring check rockets. By individually recording each pixels dynamic bracket you still only have to record 10 bits per photosite but where those bits get placed in the spectrum can mean we start to see exponential gains in latitude. Can't wait to try it out on the Phantoms think we're getting some in to test late next week.

 

[I Bloom]

Really guys take a look at the Vision Research white paper it has an excellent feature that slides the recording bracket for a photosite based on the exposure of the pixel ie its designed to record and gather valuable data for flaring check rockets. By individually recording each pixels dynamic bracket you still only have to record 10 bits per photosite but where those bits get placed in the spectrum can mean we start to see exponential gains in latitude. Can't wait to try it out on the Phantoms think we're getting some in to test late next week.

Where is the white paper?

 

[Jeff Brue]

http://www.visionresearch.com/uploads/docs/Documents/WP_Auto_Exposure_or_EDR.pdf

 

[Gavin Greenwalt]

Perhaps in the future this could also be achieved optically with photochromatic pixel filters.

Apply a logrithmic curve to the ND factor of each photo site.

Of course these unobtanium photochromatic NDs would also need to be able to transition between 'exposures' in at least 1/48th of a second.

So I'm reading this white paper but I'm still not sure how visionresearch is doing this. Is it just adjusting the exposure time for each pixel well? So it continues to expose a pixel well until it begins to approach a specific maximum value and then stops and saves that value almost like a colorfuse?

Wouldn't that result in a noticeable mixed bag of motion blur for each pixel if shooting with a >45* shutter? (Which wouldn't be a problem at all for HDR probes, but I'm curious about using it in a conventional shooting scenario.)