Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Its time to introduce a HDRI workflow

im.thatoneguy said:
:umm: ....
No. No. Noononononono.

This is like doing this mathematically:

(((36-8) + 36 + (28+16))/3)

HDRI isn't magic. It doesn't need multiple exposures. Unless you map it to real-world values like in a radiance file it's a plain old vanilla image.
Click to expand...

Yes, you do need multiple exposures if you are trying to capture an image with DR that exceeds that of the sensor you have at your disposal.

What most people think of "HDR" is actually a "tone mapped" image. They are not one and the same. You can tonemap any image with more than 1 stop of information.
Click to expand...

My point was simply that with RAW files, it's easier to dig more detail out of the image before clipping and use it to good effect. I do it all the time with still photography and tone mapping software, and it works, including the "faux" HDR method of exporting multiple exposures from RAW files and tone mapping them. You can achieve more realistic detail with less noise with that workflow than any fiddling you can do with levels and curves.
 
Buck Forester said:
Using terms like "surreal" and "range far beyond what the human eye can see" is exactly why I don't like 99.9% of the HDR images I see. They don't look real. Some people dig them, I find them aesthetically displeasing. Very unnatural.
Click to expand...

Don't dismiss it because a few people use it in a way you don't like.
 
But used "correctly" in your context, meaning, aiming for as close to reality as possible, tone mapping can achieve the exact opposite of what you describe: shadow and highlight detail that is otherwise completely lost due to the limitations of film and sensors.
Click to expand...

Just to be clearo, I have no el problemo with HDR-o. Every once in a while I'll see a tone-mapped or HDR image that looks very good. But it's just very rare. I'm guessing it's because it's relatively new so people are taking it to the extreme and thinking it looks "cool", which it does in some applications, but for nature shots I see a lot of very unrealistically tweaked images through the process. All this stuff is definitely the way of the future, no denying that. These digital chips will just keeping getting better and better in every way.
 
How about just a macro that brackets and combines the HDRI in-camera for you? The Red One is mostly a computer to start with. I imagine the Scarlet will be similar.

At the very least it could bracket the frames as it shoots by changing the exposure time (it's an electronic shutter already so why not?) Just record them to flash and let the user do the combining later with software.

Considering the frame rates Scarlet is rated for, you could shoot 5 bracket images and still get 24p. There would be some temporal artifacts but damn... that would be pretty neat.

And for still stuff you could just hit the button once and get 5+ images in one burst. Wow. Yeah. I like this HDRI idea. It could be very useful.
 
What's ironic is once an HDR image has been tone mapped it's no longer HDR... that's the whole point of tonemapping to reduce the dynamic range.

So if you're talking about a tone mapped image you're not talking about an HDR image and vice versa.

---

Bry2k I'm not certain what software you're using for your tone mapping but I would be pretty confident that it involves an intermediate step where it takes your 3 'exposures' exported from the raw file and recompiles a linear image before applying the tone mapping. Meaning you could kipthe processing steps of exporting and importing 3 images and simply import the flat RAW -> RGB image. Which is why I called it a magicians hand waving. The step of exporting 3 images and recompiling 3 images back into one isn't creating any data and if your HDR multi-exposure tools are very good ideally they'll exactly reproduce the flat 12bit linear RAW file.

---

On a side note: I just spent 2 weeks in NZ and shot everything bracketed. Even with a tripod of a wideshot you can't just compile an HDR image in almost every situation. Even the slightest breeze moves all the grass and and you get a ghosted mess. Even 2 or 3 pixels off in fine detail is going to wreak havoc on your picture.
 
ALMOST PERFECT HDR HD VIDEO is possible today.

It will take some serious money but all the pieces are there and you DON"T have to build some fantasy camera. I don't know much about video but played with a few HDR stills. Perfect HDR HD video is not possible becuase we can't stop time and we need mutiple exposures, but its easy to fake becuase of high frame rates.

I believe that I can deliver a fully working proof of concept system very quickly.

Here are the issues for Almost perfect HDR HD video
1) What can a human eye see to set boundries of the dyanmic range needed.
2) Can we display it on a monitor?
3) Its there a camera that will capture it?
4) Can we capture enough data?
5) Can we process the data to make Almost Perfect HDR HD?
6) Can we reach the goal of Almost Perfect HDR HD Video with all the loss issues and supply it to the masses.

This system can

1) What can a human eye see to set boundries of the dyanmic range needed.

ANSWER: I think human eye can see 1:100,000 contrast(feel free to correct this number if some one as a better answer)

2) Can we display it on a monitor?

Answer: Yes -- For editing: 37" 200,000:1 contrast 1920x1080 16bit 3000cd/m2 brightness .015cd/m2 black
For consumers: Off the shelf flat screens will go to 30,000:1 contrast decent bit depth and rez.

3) Its there a camera that will capture it?
I think so, there are possibilies:
A RED camera at 120fps and taking 4 different exposures to get 4 different exposure videos at 30fps at x bits. I really only need 2 exposures-- so if I lose some frame rate, bit depth or exposure do to sensor limits - I think I pretty safe. (The high frame rate solves the Freeze time issue). I think that Red can set their sensor to do it becuase there is a camers/sensor that will do 32bit with a dynamic range with 26 f-stops of exposure at very high pixel count. It not a video camera in the normal sense.

Another way:Is use High speed strobes for every other frame or every 4th frame to get the exposure difference. There some laws of phyics that come into play but seems a reasonable idea.

5) Can we process the data to make Almost Perfect HDR HD?
YES: I can show proof of hugh HDR with Still photos with only 2 shots. A true HDR encoded video with full details of the encoding process that takes care of the frame motion issues and tonal issues. Plus a better formula for encoding the HDR that should basically plug in. And another piece that lets you select what important in a frame for the HDR encoding that even simpler than photoshops background mask. They are simple programs, papers or demos that have been around for awaile that no one knows that to do with becuase they never had all the pieces of the puzzle to make something useful. They can't see the forest from the trees.

6) Can we reach the goal of Almost Perfect HDR HD Video with all the loss issues and supply it to the masses.
Answer: I believe it can be done pretty easy, becuase even with all the very complex techinical limitations and losses (aka frame rate, bit depth, camera limits etc..) people bring up.

Getting to HD rez at HDR that a person can see on a monitor they can afford doesn't seem so far off.

I have a lot of the pieces for proof of concept and know where to find the people that developed all the pieces.

I look forward to the reply's to see what I have missed.
 
If you want 3 exposure blended HDR video frames, the best you could do just now is shot a 3 frame burst at 120fps every 1/24 (ish) of a second. There are still temporal effects with that, but that minimises them.

It also gives you an effective shutter speed of somewhere between 1/120 and 1/40 depending on how smart the software is, and how much processing time you are willing to throw at it.
Something calculating motion vectors and interpretating the bracketed frames back and forth might give you something that would be usable for subjects that do not move too fast and where you are prepared to accept maybe 720HD quality.

Maybe?
 
Ok guys, this is really getting out of hands...

A) If You are posting in this thread, please pay attention to what Sandeep was referring to. He is clearly after HDRI solution targeted towards cinematography (not still photography, neither CGI). And he is after something realistic. (Sandeep correct me if I am wrong)

B) As much as I have been exchanging few "heated" posts with HD Bubaloo, he is right about one thing - no one can FREEZE time. And what (most of) You are referring to (using higher frame rates to bracket the capture) will simply not work in any real situation (with the exception of shooting mechanical objects in a studio environment, where nothing moves).

The are ONLY TWO option to do this right:

1) Use of new generation of sensors with much higher dynamic range. ARRI has already one such a sensor inside the ARRIFLEX D-21. But for most this is still year or two away and for now very expensive. And there will be still a limitations on the latitude (DR) of the sensor...

2) Use 2 or more sensors capturing light from the SAME LENS and at the SAME TIME inside the SAME CAMERA, while separating the exposure levels with the use of ND filter (cheaper and easier), or sensors with different sensitivity to light (much more expensive and complex). This way one can shoot any frame rate (supported by the sensors) and ALL frames will have higher DR.

The key to remember is not to "fill" the 32-bits of OpenEXR format (which is great for post and CGI, but useless for capture of motion pictures), but to extend the current DR of digital negatives to be equal, or greater then what 35 mm film gives us today. Only understanding this one could create TODAY a realistic solution that does not require "freezing" time and gives You full freedom of camera (and subject) motion. For the DOP nothing will change with the little exception of needing a little more light (which can be compensated by faster lenses).

Anyone from RED engineers could comment on this a little? Graeme?
 
NoahK said:
Whoa easy- just pointing out the alternative view brother.
Click to expand...

I just find it amusing that there's an "alternative view" to wanting more dynamic range. :sarcasm:

Buck Forester said:
Using terms like "surreal" and "range far beyond what the human eye can see" is exactly why I don't like 99.9% of the HDR images I see. They don't look real.
Click to expand...

So don't use it that way. Other people can do what they like.

Peter Majtan said:
The key to remember is not to "fill" the 32-bits of OpenEXR format (which is great for post and CGI, but useless for capture of motion pictures), but to extend the current DR of digital negatives to be equal, or greater then what 35 mm film gives us today.
Click to expand...

OK, but why dismiss potentially useful vfx applications? The technology probably wouldn't even be where it is now if people in the vfx field hadn't pushed it.
 
Another possible solution could be high frame rates combined with some kind of software post processing to fix areas with motion. Maybe some kind of optical flow solution would work. What if you shot with a really high shutter speed to eliminate any motion blur and then use software that finds the moving areas and smooths them out by adding appropriate motion blur? The areas in motion probably wouldn't have much dynamic range but they would be somewhat blurred anyway. This obviously wouldn't be a technique you would shoot an entire movie with but it could be interesting for special shots.
 
kmikami said:
OK, but why dismiss potentially useful vfx applications? The technology probably wouldn't even be where it is now if people in the vfx field hadn't pushed it.
Click to expand...

Mind You - I have worked in this industry mainly as VFX supervisor and have supervised few multi-million $ features earning few awards along the way.

I know where You coming from and I am not dismissing VFX applications at all. I am nearly pointing out that when post and VFX refers to HDRI it almost always is OpenEXR. OpenEXR format if unsuitable for motion picture capture in the field due to its high processing overheads... You will be pushing the current technology to even keep up with 24 fps, forget anything faster.

The truth is that we don't need the full range of OpenEXR standard to match or slightly improve the latitude of 35 mm film. 16-bit is sufficient for that and much easier to handle by in-camera electronics...

Of course You could then convert the RAW files to OpenEXR for post if You want...

But keep in mind that having higher dynamic range will make keying (green/blue screen) that much harder in the same way film grain adds to this problem. But that is a topic for another thread - feel free to start one...
 
Sycophant said:
Fine for a relatively still shot, but if there's much movement, or a camera move then it's going to look pretty funky by the time you start to build the HDRI tone-map.
Click to expand...

http://www.youtube.com/watch?v=vJs3x4BALsw
http://www.youtube.com/watch?v=0mF5jWtYh6o
http://www.youtube.com/watch?v=0U3HZJarZ30

These aren't the only examples either.

Whether this is "funky" or not is of course difficult to tell through the mush of GooTube (frankly, milapse needs to migrate to vimeo or elsewhere -- he is much too good for YouToob).

Personally, I would not expect the result, even at the pixel level, to be objectionable (given the HDR imagery is itself fine). Moving subjects would be motion-blurred as expected. In any case, there no reason to completely dismiss an exposure bracketing option for RED camera firmware. Difficulty in coding it would be a strong function of the hardware, and current software architecture (file formats and so on).

While they are thinking about that, here is my suggested improvement (if it doesn't exist already): moving average frame integration, where the current frame is the average of the last N frames. This is a common feature of better CCTV security cameras, and lets one image into seriously low levels of illumination. If the QE on the RED's sensors are about that of your typical SD video camera, then the larger pixels on the RED should give a distinct advantage (and the advantage would scale as you down-sized to 1080, 720 or even SD).
 
Timelapse is awfully different from real motion though!

Also, while it seems like auto bracketing would be fairly easy to program into the camera, I wonder how fast the auto exposure can control the iris. If the iris can't physically change every frame at 120fps then it obviously won't work.
 
Peter how does more DR make greenscreen more difficult?

If anything wouldn't it keep the screen from losing saturation if it's a bit too hot. As well as keep light clothing, skin and hair from blowing out. I've had more problems over the years with too little DR on FX shots, I can't remember a problem with having too much. :D

Or do you just mean bracketed shooting would create too many artifacts?
 
kmikami said:
Timelapse is awfully different from real motion though!
Click to expand...

I'm not sure there would be much of a difference, even at 1:1 time. I guess the only definitive answer is if someone tries the experiment at normal video/film rates. Difficult to do without some camera support though.

Also, while it seems like auto bracketing would be fairly easy to program into the camera, I wonder how fast the auto exposure can control the iris. If the iris can't physically change every frame at 120fps then it obviously won't work.
Click to expand...

I would speculate that the aperture mechanism wouldn't survive very long if could be cycled at that rate. The differences in background effects as the aperture was modulated would probably make it a show-stopper anyways. Easier would be to chop the frame collection interval into segments: 1/8 + 1 + 8, for a -3EV, 0, +3EV exposure, and HDR the trio in post.

If after trying this there were some strange stroboscopic/blurring effects, I'd suggest randomizing the bracketing to try and break up any patterns.

How open is the RED hardware, anyways? Can anyone code up/modify firmware, or does RED reserve that sort of thing to themselves?
 
Yes I started this thread in relation to cinematography solutions.
Still photography solutions and workflows exists only because the sensors are not having the required dynamic range. (Lin vs log).
Solutions are always a work around for compensating what can be achieved.
Tone mapping to 8bit is a computational and display solution. for now
Not liking the outputted tone map is an artistic matter.

I'm only trying to push for better possibilities at the DI stage, (if needed).
Truth is, you cannot produce whats not there in the first place.
And its certainly not a workaround for bad lighting.

I know that relay lenses were looked into at siggraph07.
The future is near, photography as we know could also change.

I think all these ideas just raise the awareness as to what users would like and perhaps JIM and crew are listening on this to see what more can be achieved to push the envelope.
I hope no one was hurt in the word bashing. ;)

stereoscopic is happening, 24 frm is also a subject of concern today in relation to panning stereo images

let move forward with HDRI.

put your ideas out there and lets get the ball rolling.
 
im.thatoneguy said:
Peter how does more DR make greenscreen more difficult?

If anything wouldn't it keep the screen from losing saturation if it's a bit too hot. As well as keep light clothing, skin and hair from blowing out. I've had more problems over the years with too little DR on FX shots, I can't remember a problem with having too much. :D

Or do you just mean bracketed shooting would create too many artifacts?
Click to expand...

Having more DR results in having that much more details. I had the pleasure to process some green-screen shot on the ARRIFLEX D-20, which already had much greater DR then "normal" video cameras. What normally result in a easy key turned out taking twice as long and in the end have to precise-garbage-matte all shots. The higher DR did help with keying stuff like hair, or fluffy fabric thou.

In the end I guess You win some, You loose some...
 
You must log in or register to reply here.
Back
Top