15 to 20 stop range with two RED ONE (tm)

[Dan Hudgins]

Fixture of many uses...

Fixture of many uses...

Thanks for sharing these ideas in such a public forum. Before reading this, I had never thought about this possibility.

If two RED ONE (tm) users want to team up and make a two camera fixture there are many ways it can be used, even for front projection with a single camera on it...

 

[dewaldaukema]

Original uncompressed 16 bit Tiff...

 

[Craig Ryan]

I've always thought HDR cinematography would be great for the new wave of video games being adapted to cinema.

 

[dewaldaukema]

Used lossless compression at 14 bit in the 3 x D3 originated images though...

 

[Dan Hudgins]

Camera used?

Camera used?

Used lossless compression at 14 bit in the 3 x D3 originated images though...

You used the Nikon D3 in RAW mode?

As you know RED ONE (tm) uses compression now since the RAW port is not an option(?).

I was speaking in general about the problem of displaying an image with high binary brightness segments on a display with a lower number brightness steps. If you dither pixels in a high resolution image with 20 stops into an 8 bit display at high resolution and take a few steps back, your eye averages the pixels that are one binary step up and one binary step down to make an intermediate shade. Depending on the viewing distance your eye groups four to nine pixels into a single image tone.

If you compress and image the brightness and color tone of individual pixels is no longer as close in definition to the true center value from the subject and may end up closer to one of the binary levels, when you look at the image from a distance some tone contour may get lost.

When you view frames at sound speed your eye averages the brightness and color over maybe 3 frames so the nine pixel groups become an average of about 27 pixels over 3 frames.

If you compress the images frame to frame the tones are held for more than one frame and cannot average over time, they just twinkle at a lower frame rate. When you compress in a frame, soft focus areas may get averaged in segments larger than a single pixel.

Although film has many problems, it seems the best way to exhibit uncompressed Digital images of long tonal scale right now, since the DCI seems to be a compressed format. This would depend on how the uncompressed images get onto the film though since the film recorder may have its own limitations.

If you source the images from two RED ONE (tm) cameras with different exposure you get some chaos in the compression and that helps the original tones in the subject come through when the images are combined. The fixed pattern noise in each camera will also be different, which helps also.

 

[Jeff Coatney]

Been on my mind for years- thought about getting a patent but decided not to- here is the better way:

Build a new camera with 2 cmos sensors, and a beam splitter (in the place where a viewfinder prism would go on a film camera). Lay each sensor on the back of the block (like a 3-chip RGB block)- computer aligned to match pixels. Record in sync, offload to 2 separate pieces of media (one from the "highlight" sensor, one for the "shadow" sensor- use something like Red Alert "HDR" to process the footage (flip "highlight" sensor on the H axis, tone map and export dpx). and POW! You have at least 17+ stops with great mid-tones as well (The key to this is just nailing the beam splitter to have the proper light transmission values)

Please send me a bone if you use this method! I have diagrams as well.

-Hunter

Yes, but we're back to a $250,000 camera (or cameras, since we're essentially talking about putting two of them in one box) so one must follow this great idea with the question: does the market demand for such a camera justify the huge amount of R&D expense? (my est. is three times that of the RED ONE to date) and the enormous technical hurdles involved in manufacturing to such precise tolerances. Actually, once the R&D costs are amortized into the unit costs, we might be talking about a $400,000 to $500,000 camera and that is rental business model territory. What's the market for such a camera? How many camera days per year would such a camera be used? These are questions I don't know, without research. Again a wonderful idea, but it is the balancing of the other factors in the equation that makes the hardware possible. Of course, breakthroughs in technology in any one of these areas could bring it closer to reality, but no one knows when that will happen.

 

[Michael Thornton]

What is the best software or application for doing software HDR or HDRI?

Please make a list.

Thnx

Tek

 

[Bing Bailey]

if by beam splitter you mean a prism of some kind then you are toast. the reason we can use any pl mount 35mm lens is because of the flat single cmos plane. you introduce a prism into that and suddenly you can't have 35mm lenses just like with 3 chip cameras. at least not without an adaptor and that leaves us all back at square one

 

[Pawel Achtel]

if by beam splitter you mean a prism of some kind then you are toast. the reason we can use any pl mount 35mm lens is because of the flat single cmos plane. you introduce a prism into that and suddenly you can't have 35mm lenses just like with 3 chip cameras. at least not without an adaptor and that leaves us all back at square one

No. The beam splitter in HDR does not split different colours, it splits different intensities. You can still use PL lenses, which focus all colours on a single image plane.

 

[hunterrichards]

Jeff-

I think it would be a very affordable solution as it doesn't involve new chips or new technology. All of the components already exists and are in use at this time, just not in this way. (Video taps/ OVF's/ Live Preview/ AF sensors in Dslr's)

Darkkelt-

I am not talking about a 3 chip system, but 2 chip. If your talking to Dan, I am a little confused about separating RGB channels too.
A simple beam splitter like I am talking about doesn't separate color into RGB, It just passes an image through at a certain percentage and reflects an image at a certain percentage less, think Optical viewfinder. (Image on film and Image on Ground glass) Same technology.

 

[Dan Hudgins]

Stay with the plan...

Stay with the plan...

if by beam splitter you mean a prism of some kind then you are toast. the reason we can use any pl mount 35mm lens is because of the flat single cmos plane. you introduce a prism into that and suddenly you can't have 35mm lenses just like with 3 chip cameras. at least not without an adaptor and that leaves us all back at square one

Trying to keep people on the plan, the plan was to get a longer tonal range when you need it for "free" with out any "R&D" or other nonsense, here is the plan:

1) Use your 3D fixture or front projection rig to mount the two RED ONE (tm) cameras you already own.

2) Put a 10%-90% thin beam splitter (plus IR and ND filters) into the stereo 3D fixture and align the nodal points allowing for the 1/3 thickness (a few mm) of the beam splitter. If the beem splitter in IN FRONT of the camera lenses it adds little in the way of aberation to subjects at normal distances because the cone angle in front of the lens is small. If you make a 2 chip camera there is no R&D required, just use two cameras and remount the sensor boards on a new block, if you put the beam splitter behind a single lens you use a Pellicle Beamsplitter, here is a link,

http://www.newport.com/Pellicle-Bea...catalog.aspx?gclid=CPWBrKyX5pMCFRKhxgod-ztoWg

They are so thin they do not add noticable aberration.

When you use the beam splitter in front of the two matched camera lenses you can use a thicker beam slpitter (semi-silver mirror) if it is coated on the back side.

3) Fuse the two sets of frames to make the extended tonal range image, you can use my "freeish" program or anything else that can hold a half fade. The two input frame sets should be preped by runing a curve on them so that there is not a sharp clip to the image highlights or shadows. The output result set is then run through an S curve to get film like lower contrast in the highlights and shadows. This can be done with my CC or with other (free?) programs that can adjust curves, levels and gamma.

Why would you want this, and how often would you use it? Reflections on subjects can be many stops higher than white subjects in daylight shooting. To avoid the video look, you can use two RED ONE (tm) cameras and extend the tonal range by underexposing one quite a bit, and over exposing the other a little to get more shadow detail. You need two lenses that are the same or about the same, and you need to adjust the two image sets to register by resizing them and cropping a little to get back to 4K size for the final combo files. It is a low cost option you can use when you want to to get better results, if you have two cameras and a hard lighting shot to get, why not use both cameras?

A) You can use PL mount lenses if the beam splitter is IN FRONT of the lenses. You can make a 2, 3, 4, 6, or 9 camera setup this way without problems and take the cameras out of the fixture for other uses. No "R&D" cost would be required, it is a simple way to get the thing done. Normal PL mount lenses or PL remounted relay optics can be used like optical printer lenses if you want to use a field lens for super wide angle prime lenses, or so you can follow focus easer.

B) Extending the camera tone scale allows for more natural shooting under high contrast lighting. It gives you more latitude and smooth roll off in the highlights that is less video like, along with less noise and more detail in the dark areas without having to "crush the blacks". It effectively doubles the area of the pixels and gives a more 65mm look to the images, as you know the larger the film area the more tone colors it can resolve, so more sensor area is similar.

Edison said there are three steps in the life of an invention.

1) People tell you your idea can't work or is impractical.
2) People tell you that your idea is so simple you have done nothing.
3) People claim to have invented it and you stole their idea.

I think we are at step 1... maybe all three...

 

[hunterrichards]

Word.

 

[Darren Orange]

Edison said there are three steps in the life of an invention.

1) People tell you your idea can't work or is impractical.
2) People tell you that your idea is so simple you have done nothing.
3) People claim to have invented it and you stole their idea.

I think we are at step 1... maybe all three...

Very nice...

 

[george96321]

double exposer with alternate frames

double exposer with alternate frames

Let me throw another idea into the discussion that could be implemented in software and may not require any hardware changes:
Shooting 48fps or faster and use different exposures for alternate frames. The motion effects could be minimized also in software by using motion detection compensation.
I don't know if this is old news - it just crossed my mind. It seems simple, makes me wonder why it hasn't been done yet.

 

[Andrew M.]

Let me throw another idea into the discussion that could be implemented in software and may not require any hardware changes:
Shooting 48fps or faster and use different exposures for alternate frames. The motion effects could be minimized also in software by using motion detection compensation.
I don't know if this is old news - it just crossed my mind. It seems simple, makes me wonder why it hasn't been done yet.

I am pushing for this feature with RED team since long time.
So far we got 4 features in the shutter submenu for phase, fine adjustment, relative, and other things but not the HDR bracketing.

This weekend we shot prove of concept of HDR bracketing with Bro Anansi using two genlocked cameras. One RED was set to 344 deg and the second to 22 deg. After I loaded the test footage in to the AE and merged two alternating frames together we got very good results.
First observation, you have to shoot at 48fps if you have fast moving objects in the frames.

Bro was running so fast (50' away or so from lenses) that the camera that was set to 344 deg and 24fps had blur equal the width of his body!!
However slow panning and normal people moving normally in front of the camera will give you very satisfying effect, even at 24fps.

We will repeat the shoot next week with shutter set to 48fps.

As to the merging the frames, very simple, load each shoot in to one single composition timeline.
Align the TC to be the same on both frames (turn on TC display, it is in effects menu)

From the effect distort group, pick the transform effect. This will allow you to overimpose one frame on top of the other and align both cameras frames precisely on the pixel level of both frames from camera A and B, use 400% magnification.

Also it will allow you to merge the frames using transparency adjustment.
This is the most simple HDR merge I ever used.

I used 16 bit tiffs to do the merge but in 8 bit space it works like this:
black = 0
saturated white = 255

(numbers represents the pixels on the histogram)

FRAME from camera A -- 0-0-0-0-40-125-220-180-243-226-114-216---
FRAME from camera B -- 20-47-14-56-67-33-190-212-255-255-255----

Camera A 2 stops under-exposed
Camera B 2 stops over-exposed

Zeros in camera A are crashed blacks, 255s in camera B are clipped whites.
Now, if you add these two frames you get double the dynamic range. You have to lower the exposure a bit in post after this add/merge of the frames or just play with the curves to get the correction to your taste.
For clean highlights you should cut out the overexposed frame from 200 to 255 pixel range value.
Overexposed details tend to blur the adjacent pixels. (spill off effect)

For 12 bit space (Mysterium bit depth) values will be from 0 to 4095

For HDR merge what works the best is logarithmic merge, this one will show the natural look (no painting like effect) but it will revile the details normally not visible by naked eye. (my favourite)

Also dividing the frame in to the zones of highlights and shadows and adjusting it separately will give you the best results.

 

[Dan Hudgins]

Great progress!

Great progress!

This weekend we shot prove of concept of HDR bracketing with Bro Anansi using two genlocked cameras. One RED was set to 344 deg and the second to 22 deg.

Did you shoot using a beam splitter (if so what kind) or with the cameras side by side?

I think one would get better results with two cameras in phase but different shutter angle, than having the exposures one after the other on fast motion in a single camera, best to have shutters in phase and the same angle, but the artifacts from motion may not be too noticable. Single camera 48fps would also lose a stop exposure over using a 5%-95% beam splitter.

If you are viewing compressed images on a computer screen you may not see as many artifacts as on an uncompressed film projection.

Please post more examples, I think this is a very important development for filmmakers who relish quality!

 

[Craig Ryan]

Hmm that shot above is probably one of the more "closer to the human eye" HDR shots I've seen so far. A lot of the time HDR shots have TOO much dynamic range so it begins to look computer generated...which is cool too...but a range closer to the human eye seems more practical for motion pictures.

 

[Andrew M.]

I think one would get better results with two cameras in phase but different shutter angle

It is exactly what we did.
Two cameras in phase, side by side. No beam splitter.

On top of it we had different lenses!!!
We didn’t expect to shoot that day anything, just to test the genlock and jamsinc, since we were not sure if it will work properly, after so many people here were complaining about it.

After reading the SincIT manual for 15 minutes or so, that it doesn’t happen to me very often and checking some industry standards definition on the Internet, we had all setups done faster then expected. We decided to take few shots in spite that we had different lenses installed. Good that we did it, since for tomorrow test I know what to watch for.

I want to take a test shot from inside the room through the open door, with people walking outside and with the camera slowly panning.

Then, I will merge the material in such a way that it will simulate one camera operation.

Camera A frame 1a-2a-3a-4a-5a-6a-7a-8a-9a at shutter 344 deg and 48fps
Camera B frame 1b-2b-3b-4b-5b-6b-7b-8b-9b at shutter 22 deg and 48fps

Final input on the same timeline

1a-3a-5a-7a
2b-4b-6b-8b

Final output

1a/2b – 3a/4b – 5a/6b -7a/8b frame merged at 24 fps base speed.

From timing point of view we are merging every second odd frame from camera A with every second even frame of camera B.

Since we are taking every second frame from 48fps string of frames we have to change the timeline speed to 24fps otherwise we will have two times faster speed of action.

This will simulate exactly the situation where we are changing the shutter angle from 344 to 22 for every second frame on the same camera.

Remember, 22deg shutter angle is set on the camera B that is 2 stops underexposed and any highlights may be strobbibg at this 22 deg shutter angle. It is the camera where we will be getting all these nice texture of the sky even if it is blasting noon light up there and even on the film you will get white flat look in such condition.

Camera B will be set to 344 deg so it will be overexposed. This will give us superb blacks, see the black marble columns on the right side of the picture above, the reflection of the stripes of the far side wall in black marble was not visible by naked eye.

All highlights from this camera will be rejected in HDR merge as well the highlight blur.


If this will not give us unwanted strobbing on the highlights then this shutter bracketing HDR methode will be ready to be implemented on RED.

 

[Dan Hudgins]

Great work!

Great work!

It is exactly what we did.
Two cameras in phase, side by side. No beam splitter.

If this will not give us unwanted strobbing on the highlights then this shutter bracketing HDR methode will be ready to be implemented on RED.

There may be a problem stagegering the shutter times running a single camera at 48fps, some sensors may require a frame readout after the change to the pixel readout settings, maybe someone from the RED team can answer that for the sensor used in RED ONE (tm). One might need the sensor running at 96fps (two trash readouts, two used, for long and short) giving you maybe a max of 90 degrees on the long exposure.

At some point the lens flare cuts the dynamic range of the light to the sensors, so using multi-coated lenses with a 6 element 8 surface design or 3 element 6 surface design may give the darkest shadows. There is also a 4 element two group four surface design, but only for 100mm or longer, used in some 35mm movie projection lenses for good contrast. A good EL-Nikkor enlarger lens might be something to try, 35mm, 50mm, 75mm?

A 35mm Planar maybe, or 32mm Cooke Series II may have better contrast than a retrofocus?

The matte box should almost cut in on all sides of the frame.

I have thought of using a cemented triplet, cut down like an hour glass and painted black on the sides to only have two air to glass surfaces, that would have a fixed stop but you could change the element in the mount, it should give the lowest flare possable, if made right.

Great work, can't wait to see examples from your next tests!

 

[Andrew M.]

There are two types of HDR work, one is when you want to take 4-5 stops lit scene and stretch it to 14 or more. The second one is when you have 15 stops scene and you want to capture both ends and compress dynamic range back to screen/projector dynamic range of 8 or so.

In the first case the sensor dynamic resolution precision and lenses quality is very important.

Example:

Take the shot of blue sky or perfectly black marble surface from the picture above.
Uniform blue sky or black surface does not contain any texture.

Maybe in the different lighting condition you could see some texture on the sky left by passing jets few hours ago, or stripes on the marble pillar but blinded by the sun you will not see any texture there.

Human eye is not good in recognizing small difference of intensity, and also not too good in recognizing high contrast scenes as well, as estimated by some sources as only 6.5 stops.

The second case is the high contrast scenes of 12-16 stops.
What is important here is not so much the intensity precision and fine resolution of the steps as the number of steps.

In our experiment 12 bits (12 steps) of Mysterium sensor shifted by 4 stops between camera A and camera B is filling up whole 16 bit TIFF space, wall to wall.

Example:

HDR classic, church door looking outside on the sunny day with a few clouds.
You want to show the church alcove and side altar detail at the same time it would be nice to see these small clouds on the blue sky in the door.

Well again, human eye looking straight in to the door will not see the altar details at all, it will take a minute to see it once you turn your head and start staring at the dark of the church interior.