HDR Mode...

[Felix van Oost]

Is this #21?

I don't think so. The HDR mode was already mentioned in the list.

 

[Kyle Presley]

For stills, not motion...

 

[Andrew M.]

For stills, not motion...

both my friend, both......

 

[Michael Schrengohst]

Using HDR in practical ways.
I shoot 360 Panoramas.
I shoot hotels, motorhomes, even Audio Post houses.
Shooting 360's and trying to use additional lighting is almost
impossible. So you have to shoot with the light you have.
Most of the time I use 5 brackted images.
So in a pano that is 35 images.
I have used as many as 9. I cannot shoot one image anymore!!! LOL....

Audio Post House
http://izonemedia.net/PANO_DEMO/PA_PANO/pa_2.html

Crowne Plaza Hotel
http://izonemedia.net/PANO_DEMO/CROWNE_PLAZA/cp.html

High end MotorHome
http://izonemedia.net/PANO_DEMO/MHSRV_PANO/mhsrv.html

12" snowfall in Dallas, Texas
http://izonemedia.net/snowday/snowday.html

 

[Andrew M.]

As for those processing the same shot different ways to get HDR-like results, I had a conversation years ago with Graeme asking about this, and he'd said if you only took one original image, no need to process out to various other shots - it is all about the curves. I agree.

-mike

It is all about curves if you have excess of DR. You can’t push curves anywhere if you are taking the picture like, looking out of the church door in to the sunny sky and you want both indoor and clouds to be looking nice. (Classic example of the HDR)

I took many HDR pictures that do not look artificially pushed but yet showing amazing picture dynamic range. Whenever there is lots of lights and shadows to show HDR comes handy.

 

[David Battistella]

sorry what i meant was not just changing the ISO from the RAW file which is just metadata. Instead, actually shifting the native ISO of the sensor.

No.

I was asking:

Is HDR mode a metadata mode or is it baked into the RAW file?

David

 

[J Davis]

No.

I was asking:

Is HDR mode a metadata mode or is it baked into the RAW file?

David

This is what I was asking right back at post 12.

JJ says 'There will be an HDR Mode on EPIC and Scarlet. ' which I interpreted as a 'mode' and as on camera.

If it was manipulation of metadata he would have promoted it as a part of Redcine X. From this I conclude it is baked into the RAW file.

 

[J Davis]

BTW, the Nov 30 announcement information at this link
http://reduser.net/forum/showpost.php?p=517037&postcount=7

for scarlet 2/3 list currently reads '31. HDR Stills Mode'

is this now incorrect or do the 2/3 scarlets get HDR motion as well?

 

[Joel Kaye]

No.

I was asking:

Is HDR mode a metadata mode or is it baked into the RAW file?

David

That's the real question. We know the sensor gets what the sensor gets unless shutter or iris is adjusted. Adjusting IRIS causes DOF problems... ergo the shutter is going to get adjusted or it's some kind of on camera curve/shadow/highlight adjustment. The latter might be a workflow timesaver but wouldn't actually create an image that couldn't already be created in post right now.

If the shutter gets adjusted on the fly, captured and then baked into the raw data that would be pretty interesting.

Next question becomes - does that work on moving subjects? I don't currently understand how that could work.

 

[Richard Goodwin]

Rather than tone mapping or post techniques it could be done at the sensor and ADC circuit level.

possibly this is the way it is done (bear with me here):

A technique that would work would be to take multiple samples of each sensor “exposure” for successively increasing times. This would greatly reduce the elapsed time of exposure but let you have multiple data samples of increasing density.

What I am saying is that imagine a single exposure taken by the sensor (what we think of as a frame). Now that exposure lasts for a certain amount of time. The analog to digital converter usually takes a single sample of that exposure which results in a “frame”. But, it could take several samples of that single sensor exposure, each one increasing in length. In this case each sample would start at the SAME time (time ZERO) but last successively longer. So for an exposure you might take three samples.

Here is an example imagining how you might get three HDR frames from the same exposure. The sensor begins to “expose”. Lets call that time ZERO. Now the digital to analog converter starts recording THREE samples of the same exposure at time ZERO. The first sample is taken for 100th of a second then stored, the second is taken for a 75th of a second then stored, and the third for 50th of a second the stored. As they all started at the SAME time the total elapsed exposure time was only 30th of a second. At the end these three samplings could be combined into one to produce an HDR frame. Because you have three different exposure times of the SAME frame).

Using a technique like that will produce three (or more images) of varying density but will not require radical changes in exposure time. Hence, you don’t need to take 3 separate frames, and even motion sequences will work because you can still do cinema FPS rates AND avoid significant mis-matches in the motion in the frames.

Anyways, that is my shot in the dark. Maybe that is why the ASIC is 29 layers ☺

 

[Joel Kaye]

Rather than tone mapping or post techniques it could be done at the sensor and ADC circuit level.

My understanding is RED doesn't do any analog gain amplification. I think they tested and found no benefit. Unless someone like Graeme posts differently then I'm pretty sure the sensor gets what it gets and it's really, really smart software (or ASIC) from there. The ASIC could be tone mapping. I can see that... that would be useful.

 

[Zhibo Lai]

This came to me while I was in the shower. This might be a completely whacked out idea, but thought I throw it out there.

I'm no HDR expert, but my understanding is any trick involving bracketing of shutter speed will result in ghosting of some sort. The more motion, the more ghosting. Thus, I throw out shutter bracketing.

What if the HDR trick is adding a second brain, thus 2 sensors. Some possibilities here:
1 - The main camera feeds an unprocessed (perhaps analog signal) with the full dynamic range the camera can see and splits this between the 2 brains via a data link, and this is bracketed to each sensor.
2 - Light splitting is done optically through a lens splitter adaptor, one goes to the main camera, the other passes through an ND filter to the 2nd brain.
3 - same idea as #1, but with a special HDR module, instead of brain, that is an MX sensor tuned specially to capture a higher or lower range of a regular brain.

I don't know if this is even possible, but it would be cool if something like this could actually be pulled off. Want more DR, just add another brain, the more brains you add, the more DR you can get, limited by number of times a signal can be split without noticeable quality loss.

This is much the same concept as in CPU speed. Instead of cramming more circuits into a single chip to yield faster machines, they switched to using parallel processing so you get faster machines by adding more CPUs. Do the same with sensors to yield greater DR, or maybe even more resolution.

My final guess, and probably more realistic of them all is: Graeme has invented a new algorithm that pulls extra DR from a 4D super FLUT curve off the MX sensor to bring us HDR. Not extra hardware or optics needed, just Graeme's super FLUT sauce.

 

[Richard Goodwin]

My understanding is RED doesn't do any analog gain amplification...

I'm not suggesting any change in the gain on the sensor itself. I'm suggesting that they sample the sensor for longer and longer intervals. But those samples start at the same time. So the work is done in the ADC and related circuits that take data from the sensor. The sensor sensitivity (gain) does not need to be varied at all. All that is varied (in this case increased) are the lengths of sample captured

 

[Joel Kaye]

All that is varied (in this case increased) are the lengths of sample captured

Hmm... I think you just went past my pay grade. LOL. So you believe sampling the sensor for longer and shorter times would alter the amount of visual data the sensor is getting and those samples could be combined to extend dynamic range? I just don't have enough knowledge of how sensors capture data to have any clue as to whether that would work.

 

[Richard Goodwin]

Hmm... I think you just went past my pay grade. LOL. So you believe sampling the sensor for longer and shorter times would alter the amount of visual data the sensor is getting and those samples could be combined to extend dynamic range?

Just having fun speculating.

The visual data would be the same, only the density (the exposure times) would be different. The key here is all the exposure start at the same time (they are coming from the same sensor and here exposure is just what I am calling the "sampling").

Here is an imagined scenario of the idea: It would be like you had three pieces of perfectly clear film stacked behind one another in a camera. You open the shutter and all three are being exposed to the same image. Now after 100th of a second you pull the first sheet of film away, then the second after 75th, and the third after a 30th. So you would have three exposures of the same moment but of different exposure times (hence different densities). Of course in this imagined world the film does not blur when you pull it away and it stops exposing etc. But, that is a physical analogy.

I'm just theorizing here so anything is possible I guess more technically it would depend on the ability of the sensor to recharge fast enough and the sampling and processing circuitry to be fast enough.

 

[C_Skaff]

sorry what i meant was not just changing the ISO from the RAW file which is just metadata. Instead, actually shifting the native ISO of the sensor.

oh, that made me think of a thing...

Is it even possible to change the true native ISO on a sensor?
I don't shoot still cameras, but my friend swares that on his Canon he can change the ISO in camera and NOT in post, and that that is actually changing the sensor not any Metadata... but I controdict, Red changes the metadata and NOT the sensor itself because that would be physicly impossible. It would be like turning a lamp up higher then Max. A sensor i linear and can't be changed...
Or have I got it wrong?

 

[C_Skaff]

HDR means High Dynamic Range to us... it does not mean tone-mapping. Speculate all you want here. We aren't giving out the specifics of how it works until it is released.

Jim

HDR... doesn't that mean HD-Ready?
I though it was way more then HD, even more then 4K

 

[Noah Yuan-Vogel]

I'm suggesting that they sample the sensor for longer and longer intervals. But those samples start at the same time.

Seems to me the motion artifacting could be significant. to significantly extend dr youd need samples that are quite different in exposure time, think about what would happen to a bright object moving in the frame... for the longer exposure, it might blow out, and for the shorter exposure, a good part of its motion blur would not be captured by the shorter exposure, youd likely get changing exposure and different amounts of highlight detail in one object's motion blur. every frame might start looking a bit like multi-curtain shutter shots or like those still images with a 2nd curtain flash sync (am i describing that correctly?).

 

[Noah Yuan-Vogel]

oh, that made me think of a thing...

Is it even possible to change the true native ISO on a sensor?
I don't shoot still cameras, but my friend swares that on his Canon he can change the ISO in camera and NOT in post, and that that is actually changing the sensor not any Metadata... but I controdict, Red changes the metadata and NOT the sensor itself because that would be physicly impossible. It would be like turning a lamp up higher then Max. A sensor i linear and can't be changed...
Or have I got it wrong?

Canon definitely does not use metadata iso in its cameras. Red definitely does it its Red One.

 

[Gavin Greenwalt]

Let's say the sensor can shoot 160 fps in 3k mode. That means in 1/48th of a second it can shoot 3 frames. Capture those 3 samples at a 360* shutter. Now add them together. Voila. +1.6 stops of dynamic range. For increased dynamic range. You would need to reduce the combined shutter speed to allow for more samples.

Even a small change such as shooting 24fps @ 1/40th shutter would get you all the way up to +2 stops of dynamic range.

The huge assumption with such a system would be that the gap between 360* shutter frames is extremely small. Otherwise you would notice the 3 frames as 3 seperate samples and not one continuous exposure.

This seems like the most likely system since it doesn't require any terribly sophisticated software. It just needs two things:

1) The ability to shoot bursts of 360* shutter frames in little seperate bursts which add up to a user defined exposure length instead of continuously shooting.

2) A system which applies a multiplication to the frame buffer before adding together (if you need to keep the range between 0-1.0) or a system which can hold a floating point frame buffer with values > 1.0.

Assuming FLUT can handle floating point frame buffers > 1 for the RGB output processing I imagine this makes Scarlet/Epic capable of handling this.


We already know that Scarlet can hold at least a handful of frames in memory before writing to disk since it has a burst mode. So all of the components are there. High frame rate sensor, floating point space image processor and buffer capable of holding multiple frames in memory.