Hdr

[shoook]

One of the most impressive modern art forms is the HDR photograph. I know some cameras have a high dynamic range option and many of us use 35mm adapters to get that film look but I've never heard of someone applying the HDR techniques of photography to film mediums. Photographers use Photoshop and HDR generating software such as Photomatix to combine several images and achieve the HDR look, but are there any tools out there currently being used to get the same results on HD video?

http://farm1.static.flickr.com/69/182191565_0537107963_b.jpg

http://farm3.static.flickr.com/2119/2049233526_358678b16e_b.jpg

[wgzn]

as far as i have always understood, the HDR stuff is achieved by combining multiple exposures of the same shot. dont really know how youd achieve that in video without multiple sensors or some other in-camera situation...

[DM_rider]

This has been brought up and discussed before. Pretty sure there's no way to do it with a video camera. I have however seen many HDR time lapses.

[David Shawl]

Here's a cool HDR time lapse my friend shot: http://www.vimeo.com/2822787

[Nick Walters]

What about over cranking, combining via HDR in post for an end result of the desired frame rate?

[Nick Walters]

With my HVX-200 I can go 48 frames....take 2 pictures per, down to 24...synch audio off another recorder....of course I know nothing about HDR, but whatever

[Postmaster]

But you would need alternating exposure times.

The only way I can think of is a camera with 3 CMOS Bayern pattern chips behind a prisma.

Frank

[Everts]

Have you been wacthing tv, been to the movies lately ? its there , nothing knew

Batman darkknight has a scene wer he stands on some building looking down.
The city was made by 12 megapixels fotos stich together . for the high dynamic range
They did the same thing with the rush hour where they fall of the eifel tower . The whole city except the tower was made of HDR / raw photos

HDR images are especially used in composting cg/ and even real life fotos with real life action shots.
Checkout fxguide.com

[Nick Walters]

What if you used the extra frame in a 48fps setup...changed it around digitally...twice, and then formed them into an HDR...I realize this would be an amatuerish effect at best, and perhaps would introduce noise but perhaps the results would atleast look interesting, lol :)

[Everts]

This has been brought up and discussed before. Pretty sure there's no way to do it with a video camera. I have however seen many HDR time lapses.


I think those were shot with varicam' s or hpx 3000 s.
On another note that is what the DRS function is for , but for now its somewhat noisy maby it will get cleaner in the future.

[Barry_Green]

But how would you alter the exposure between frames? HDR requires you to have two different frames, shot at different exposures, and then combined in post.

One way you could do it is to shoot with two cameras through a beamsplitter prism so that they had identical fields of view. Then you'd set one camera to have a higher exposure than the other, and hope and pray that they're genlocked perfectly in phase with each other.

Doing it with just one camera, using faster frame rates, is never going to yield ideal results because -- these are motion cameras! So any motion between frames will yield "interlace"-style motion artifacts when you go to combine the frames. So even if someone built a camera that took alternating exposures (probably by adjusting the shutter speed, so every even frame is taken with a normal shutter speed and every odd frame taken with a high shutter speed to lower the exposure and capture more highlights) then when you go to combine those frames in post, there will be motion differences between them which will result in jittery edges and interlace-like artifacts.

[Nick Walters]

Well I guess what I meant was basically taking every other frame, adjusting it's curve in post, twice...one that's a bit high, and one that's a bit low...then combining them. Perhaps this would just output a file that looked exactly like the original, I don't know.

Or perhaps I'm thinking in the wrong way entirely. to avoid motion blur artifacts you shoot whatever normal frame rate, adjust the curve on each frame twice...for an original, a lower, and a higher, then combine for an HDR. But like I said perhaps that would just output something similar to the original.

[Nick Walters]

Also, to avoid motion blur artifacts, I think you could go film cam, 48fps, then change the shutter to 1/1000....this would be pretty dark stuff, but it should reduce motion blur...or am I thinking in the opposite direction?

[shoook]

Well I guess what I meant was basically taking every other frame, adjusting it's curve in post, twice...one that's a bit high, and one that's a bit low...then combining them. Perhaps this would just output a file that looked exactly like the original, I don't know.

I think this could work if you were doing a 48fps recording of something from a very wide angle that had very little movement. If this was the case the interlacing Barry is talking about wouldn't be nearly as noticeable. At least that is my theory.

[dcnblues]

Well I guess what I meant was basically taking every other frame, adjusting it's curve in post, twice...one that's a bit high, and one that's a bit low...then combining them. Perhaps this would just output a file that looked exactly like the original, I don't know.

Or perhaps I'm thinking in the wrong way entirely. to avoid motion blur artifacts you shoot whatever normal frame rate, adjust the curve on each frame twice...for an original, a lower, and a higher, then combine for an HDR. But like I said perhaps that would just output something similar to the original.

You are thinking in the wrong way entirely. The part you're not getting is the RAW data. HDR works because you combine at least three original images, all with different data / exposure.

You can't take one frame, manipulate the date into three images, and think there's any useful gain in that. You need the data of three different exposures, and combine them for the desired effect. As other's have said, for realtime motion, you'd need a beam splitter.

I actually don't think this too far off. Many cameras use internal prisms to split the light into 3 separate ccd sensors (each sensor dedicated to an individual color). What if (as Postmaster suggested) you designed a camera with 3 separate CMOS sensors (one sensor dedicated to all three colors), each dedicated to a different bracketed exposure?

You'd need to triple the camera's computing power, and storage, and might not be able to see a live output in full resolution immediately, but we're at a place where big advances are being made in the hardware. It would be pretty cool...

Alternately, you could get three less expensive cameras, design a big clunky bracket system to have them all feed off the same lens, set them to bracket exposure, and write some code to batch process the footage in photoshop. A lot of work, but doable, and could result in some AMAZING footage that would deliver tremendous bang for buck. As I said, I'd be amazed if no one out there was working on something like that right now...

What would really be fun is to design such a hypothetical rig, and take a poll of DP's. I honestly don't know if they'd hate the idea, or love it, or both. It would make most set lighting much, much less difficult, hugely increase efficiency, but I'm guessing wouldn't really help higher end results. Art sometimes doesn't react well to artificiality, and HDRI images are intrinsically artificial.

[dcnblues]

This has been brought up and discussed before. Pretty sure there's no way to do it with a video camera. I have however seen many HDR time lapses.

Actually, and to make this relevant to this forum, they're already doing it, to a lesser extent, in the HPX170. Look up DRS:


-Okay, and now for the biggie: the other image control they've added is DRS (Dynamic Range Stretching). This is basically an in-camera way to get High Dynamic Range images. It's a feature imported from the AJ-series HPX2000/HPX3000. This one feature greatly expands the appearance of dynamic range across the scene.

[Chris Adams]

Check out this article by John Galt, the Senior Vice President of Advanced Digital Imaging at Panavision’s corporate office:

http://magazine.creativecow.net/article/the-truth-about-2k-4k-the-future-of-pixels

Scroll down past all the stuff about 4K(although I highly recommend reading it) and he talks about how they are integrating HDR into their new sensor.

[shoook]

When I first heard about the DRS feature of the HPX I was really excited. Then I learned that you could only do DRS in camera mode so I just haven't used it.

What if you put two cameras side by side or stacked one ontop of the other and then set them to different exposures. Do you guys think this could achieve an HDR look?

[monday1313]

Check out this article by John Galt, the Senior Vice President of Advanced Digital Imaging at Panavision’s corporate office:

http://magazine.creativecow.net/article/the-truth-about-2k-4k-the-future-of-pixels

Scroll down past all the stuff about 4K(although I highly recommend reading it) and he talks about how they are integrating HDR into their new sensor.

thanks Chris, you saved me the trouble of finding that link again to post it myself...

[Nick Walters]

When I first heard about the DRS feature of the HPX I was really excited. Then I learned that you could only do DRS in camera mode so I just haven't used it.

What if you put two cameras side by side or stacked one ontop of the other and then set them to different exposures. Do you guys think this could achieve an HDR look?

I know photoshop can automcatically stitch stuff together for HDR, so perhaps something could be done which would look smoother then the time lapse stuff.....I LOVE the time lapse HDR, but if it were smooth motion, it's be so psychedelic :Drogar-Love(DBG):

Perhaps something in 1080, which is then cropped to 720 once the stitching is done...to remove the areas that didn't get enough data to effectively do it.

[Postmaster]

As I already said:

The only way I can think of is a camera with 3 CMOS Bayern pattern chips behind a prisma.
Would be "easy" to made. Take a 3 chip camera and replace the CCDs with CMOS Bayern chips.
Than you need a circuit that let you control different exposure for each chip and the way they blend together - et voila - there you have it.

More sophisticated cameras may record 3 raw streams and you do the blending in post.

I think most chip and camera companys already working on HDR (like Panavision).
I predict HDR in consumer cameras in about 5 years.

That will be the next big hype, not 3D.

...well maybe 3D HDR .

Frank

[Nick Walters]

Here's a short video I did, mostly to prove to myself that the workflow was possible:
http://www.youtube.com/watch?v=zyBY01ZBo10
http://vimeo.com/4480709

basically I just setup a tripod shot, and filmed a few times at different exposures. I outputted a Tiff sequence from Premiere and trimmed everything down to exactly the number of frames I wanted, used Photomatix to make a OpenEXR HDR sequence and to align any images that went funky. Then in After Effects I motion controlled the final frames for further alignment and adjusted the levels for viewing of all the 32bit data and outputted a lossless AVI. Then in Premiere I did basic editing, addition of titles, etc....I also laid over some music from Sony Cinescore.

I was thinking of five streams/cameras at first, but then I realized that four cameras using beam splitters and a 4 way video combiner(used in security setups) could make for a pretty good setup where you could quickly see what areas were over/under/well exposed. There's a ton of post work to be done in such a setup, but it seems it's definitely possible.

The biggest problem I could see is getting the images on each camera to be framed exactly the same.....one university research team in my area used B&W sensors and a retroreflector to aid in alignment

[monday1313]

nice work, what time of day was that shot at?

[Nick Walters]

nice work, what time of day was that shot at?

Thanks, I believe I filmed around 11am-1pm in the St. Louis area. I'm actually quite sure if I would have waited for the right moment and perhaps picked a better subject the results would be much more evident....something with alot more shaded areas, alot of cumulus clouds, perhaps some lights here and there.

I'm really hoping someone with either access to a few cameras, or the cash to buy them will pick up the idea and expand on it so that motion can be introduced. The beam splitter mirrors are relatively inexpensive on ebay, and a rig could be built from wood or any material laying around.

I also believe that the further away a subject is, like say just moving clouds, the easier it could be to just forgo the building of a rig, and one could just line up the cameras close together(like say four cameras in a square pattern), and depend on Photomatix's ability to align the images, and then After Effects motion tracking for further correction....it might even be possible to do this by going full zoom and backing way up to record a more full scene....the big thing to remember though is that you always need a reference point for After Effects to base it's motion tracking on.

[pmark23]

I posted results of my tests from 2005 somewhere here (can't find it now).

Basically, the shots have to be perfectly aligned or there's major problems with solarization. If it's off even by one pixel you'll notice.

You can get away with it with still photos, but it really shows up when there's out of phase movement.

[Nick Walters]

It sounds as if you weren't doing HDR exactly, but just trying to combine a couple images to make one big well exposed one. While this is generally what we're shooting for in HDR, it doesn't exactly hit the nail on the head. One of the reasons HDR imaging programs can work well is the use of special algorithms, image alignment, and such things. I even go the extra step of saying that the images must not only be aligned, but the use of motion tracking must be used in After Effects to offset some jiggle in the frame caused by the HDR imaging softwares image alignment....as you see my results aren't perfect in my video,. but 2 seconds of solid video(48 frames)...seems to prove, atleast to me, that further testing could be done with possibly good results.



While procrastinating, I did a little though experiment, then backed it up with a real experiment.

The idea came to me while doing some HDRI (High Dynamic Range Image) work in Maya. You can read a bit about HDRI here (http://www.highpoly3d.com/writer/tutorials/hdri/hdri.htm).

It would be really cool if someone else tried this as well, particularly those of you with two DVXs, a beamsplitter, a 35mm lens adapter, and a bit of old-fashioned gumption (a machine-shop would help too).

I put the DVX in my backyard, and shot a minute of video while slowly opening then closing the iris. The camera was locked down, and nothing was moving in the background. The scene was a combination of dark shade, and bright sunshine -- a typical nightmare scenario for video.

Then I captured two stills into Vegas, one underexposed, and one overexposed. They were one on top of the other. Then I created a mask (automatically, using "Secondary CC" luminance mask) to "punch a hole" in the overexposed areas, so that the under-exposed "bright sunshine" parts showed through.

The result is that the dark bits were perfectly exposed (from the over-exposed shot), and the bright highlights were perfectly exposed (from the under-exposed shot).

There were problems with solarization on the overlapping areas (but I already knew this would happen from the "thought experiment" stage), and it took a bit of fiddling to get a good (but not perfect) match. A bit more experimentation and I think we would be able to get a perfect image.

I propose that a rig be made consisting of two DVXs focusing on a beamsplitter (prism), with a single 35mm lens at the front-end. One DVX would be configured to overexpose, and the other to underexpose. This way you capture as extreme a latitude as possible, and can "set" or "fix" the exposure in the NLE while editing. This will add another step during post-production, and ideally a plugin or separate software will need to be written to automate this.

Seeing that there are quite a few of you who are into making "35mm adapters", there's not a big leap into making a "HDRI/V" (as snappy a name as I could come up with) adapter. Unfortunately I don't have the resources.

One potential problem is syncing the camera's, but we'll only know if this is a problem when a complete rig is made. I think that natural motion-blur may make this a non-issue.

If this does work, when the HDX comes out this rig will CRUSH film. Do you hear me, film? We're coming to get you. Your days are numbered! :laugh:

[Nick Walters]

Here's another video showing a 1/2 frame drift,. by usin 48fps in a 24 timeline....just shows that a little drift won't completely ruin a piece of video:
http://www.youtube.com/watch?v=4TApJxQGsXo

[dougspice]

I feel like I'm not seeing anything in your examples that doesn't just look like normally exposed photography. Especially in the second one.

[Nick Walters]

I feel like I'm not seeing anything in your examples that doesn't just look like normally exposed photography. Especially in the second one.

They are only proof of concept videos really, and the second is definitely not HDR....it only uses the workflow to display the fact that a little bit of mis-matching in the frames during the HDR image creation doesn't have a huge negative effect on the end result.

I do have to disagree about the first video though, as I would never have been able to expose the shot that well without lights and bounce boards....perhaps the next one I do, I'll do a side by side video...one showin a just generally well exposed shot, and the other side showing the shot done with HDR.

No matter what though I won't be able to fully display the HDR capability until I can film a scene simultaneously with multiple camera's, instead of fudging it through multiple takes.

[Nick Walters]

I also wanted to point out that the first video is done without a polarizer....this is all straight through the lens.

I have some free time today, so I'll try to find a nice location that can really bring out the contrast and color in HDR.

I'll film the scene a total of five times...four at four different exposure levels, and a fifth that is just well exposed, and I'll display them side by side.

[Nick Walters]

Here's a link to a video I did comparing regular video, color corrected/enhanced regular video, Plain HDR video, and tone mapped HDR video.

I would suggest watching the HD version in full screen, and perhaps pausing to inspect the differences, as the actual core video is only 20 seconds long.....as I knew the post process would be quite long, so I tried to keep it short. Sometime in the near future I'll be trying to do a 1-2 minute long HDR video incorporating more motion.

Here's the link:
http://www.youtube.com/watch?v=4sIEFDuuaSg

[Huy Vu]

Sigh, if only we have a camera actually capable of capturing that dynamic range. The HDR with tone mapping looks very nice. The reference adjusted is actually pretty good; you got back quite a bit of details without losing too much overall contrast. I would suggest showing the videos one after the other instead of squeezing them all on the same screen, it would be nice to see high res examples.

[Nick Walters]

I thought someone would ask for this....I just woke up in the middle O the night, and I added everything to the render Q, so I should have something late tomorrow night....it would be sooner, but I have a shoot.


Sigh, if only we have a camera actually capable of capturing that dynamic range. The HDR with tone mapping looks very nice. The reference adjusted is actually pretty good; you got back quite a bit of details without losing too much overall contrast. I would suggest showing the videos one after the other instead of squeezing them all on the same screen, it would be nice to see high res examples.

[dougspice]

The HDR with tone mapping looks very nice.

Really? I think it looks by far the worst. Uninteresting and washed out. Images are supposed to have contrast!

[Nick Walters]

Really? I think it looks by far the worst. Uninteresting and washed out. Images are supposed to have contrast!

I thought the HDR without tone mapping turned out well, it was the closest of the four to what I actually saw with my own eyes

[Nick Walters]

Here's the HDR video full screen:

http://www.youtube.com/watch?v=fgSkDtMPaBk

I'm doing some experiments now on merging HDR with normal video, and should have some video to share soon......

[Nick Walters]

Really? I think it looks by far the worst. Uninteresting and washed out. Images are supposed to have contrast!

The more I work with HDR the more I understand how to incorporate it, and one major thing I've found, based in no small part by your comment, is the need to blend in the original image with the HDR.

I'll have 58 seconds of video to show in about 3 hours, in which I leave the original videos opacity at 100%, the HDR without tone mapping at quite a low opacity,. and the HDR with tone mapping at an even lower opacity....the layering of these different kinds of images leaves alot of detail in the shot, a good amount of contrast, and although many of the colors are exxagerated I think the results are quite nice.

Stay tuned for the video.....

[Nick Walters]

Here's my latest HDR video:

http://www.youtube.com/watch?v=HSmnb-JZNbg

[Nick Walters]

Another video...it's almost exactly the same as the last, but I set another layer on top near the end that increases perceived dynamic range:

http://www.youtube.com/watch?v=5PHcitPfNBk