High Dynamic Range RED

[Emery Wells]

While in 2k windowed 120fps mode the aperture would change 5 times a second (bracketed exposure essentially) to capture all available light from the brightest whites to the deepest blacks. The 5 frame sequence of exposures would be post processed to one of the many high dynamic range formats (EXR, HDR, etc).

For those of you who are not familiar with HDRI, it is an image format typically produced from 3d packages or by taking a series of bracketed exposures with a still camera and then combining them together in post to give you a super wide latitude. This can then be used to step through exposures and choose the one you'd like, tone map the image to create an image that looks closer to what the human eye sees (super bright and super dark are in visible range at the same time,) or this can be used to make a light probe which is used to apply image based lighting to cg object.

HDRI RED would allow for super color correction and exposure control,, *moving* image based lighting on cg scenes and countless other really cool things that people would think of.

Example:
Exposure 1:

Exposure2:

Exposure 3:

Tonemapped:

I win,
Emery

 

[Jay A. Kelley]

This is an incredible idea.. I can only imagine how difficult it might be, but still it would be incredible. bascially we're looking at 3 movie files, lay them into the editor and mix up the PERFECT picture.

This idea gets my vote.. It's unique for a video camera, and would allow a type of color correction not found anywhere else.

Great thinking
Jay

 

[Policar]

You can't do that with a moving camera, and capturing 5 images per second increases the data rate too much to be useable.

However, increased dynamic range is aways nice...(even if it means a washed out image most of the time.)

 

[Emery Wells]

This would not change the data rate. The camera is already capturing at 120fps... you would just be changing the exposure 5 times a second with the objective of having a 24fps HDR sequence.

Im well aware of the limitations when shooting with a moving camera. It would require post processing algorithms to delineate and interpolate data across the frames. This does need need to come from RED. There are lots of things that could be done with the sequence in post using pixel vector motion analysis, tone mapping, etc.

 

[im.thatoneguy]

I've looked into trying to hack existing cameras to do just such a thing. The problem that cropped up when I thought it through was a question of consistancy. If you have fast motion, you're going to get very very odd artifacts. The algorithm for the HDRI creation could get a person in 4 different places in the frame at once. It would think that the luminance change was from the exposure not the scene changing, and you would get strange values.

I would love for DSLRs to do this however, the closest I've found is you can usually program an automatic bracketing macro and complile the HDRI on the PC.

 

[Policar]

The trick isn't necessarily HDR (however cool it can be) but instead a sensor that not only records brightness levels at each pixel site, but also HOW LONG it takes to reach a given level. That way, values can be interpolated based on time for all 100% white pixels allowing for...well...tons of dynamic range. (And you only need to store time for white pixels, all others can be stored purely as light level.)

A contrast knob would be awesome, too, but that's more a videography thing. I had a shoot where I needed to shoot a projector in a lit room, and no matter what f-stop at which I was, I couldn't get a good exposure...always washed out.

 

[MrDorf]

The trick isn't necessarily HDR (however cool it can be) but instead a sensor that not only records brightness levels at each pixel site, but also HOW LONG it takes to reach a given level. That way, values can be interpolated based on time for all 100% white pixels allowing for...well...tons of dynamic range. (And you only need to store time for white pixels, all others can be stored purely as light level.)

Exactly, you would need a new Mysterium Sensor with a 4th channel, then it could adjust each pixel of an image based on an overall brightness over time reading. You would never have a blown out area again, because those individual pixels would "realize" they are past 100% contrast and adjust themselves down in brightness. Likewise darker pixels would "expose" slightly longer and bring up those dark areas.

You'd effectively get an HDRI image without having to capture multiple frames. You'd also never have to mess with the huge software/processing sink of dealing with offsets and interpolation to make 3 images taken at different moments into the same moment of time.