Anaglyph Previews on Web -what codec?

[Phil Bates]

Our favorite codec for the web is H.264. However, with anaglyph images, the artifacts are somewhat distracting. The effect is as if there is a textured glass sitting on the screen plane.

Here is a comparison (you'll need those fashionable red/blue 3D glasses):

H.264:
http://dl.dropbox.com/u/5981875/Anaglyph%20Compression/H.264.png


Original:
http://dl.dropbox.com/u/5981875/Anaglyph%20Compression/Original%20PJPEG.png

Any ideas on a better web friendly codec?

Thanks,
Phil

[C. Jagan Mohan]

Problem is not codec it is cardboard 3D and the logo is pulled out.

[Phil Bates]

Hi Jagan,

Thanks for responding, but I am not sure I understand. We set the logo in front of the screen plane on purpose. But the problem is not the logo, imagine it not being there.

Also, regardless of the cardboard comment, the elements still have depth due to negative parallax. The distraction is due to artifacts that seem to sit on the screen plane, especially around the lit green trees. Are you saying you don't see a difference between the top and bottom images?

Thanks,

Phil

[C. Jagan Mohan]

Is this 2D to 3D converted stuff?

[Phil Bates]

No, shot with 2 parallel REDs.

Phil

[Thor Melsted]

I think Youtube's approach is better, but more complicated and has other drawbacks, but it doesn't suffer from this problem.
There's one source file in a side-by-side compression (which means halved horizontal resolution per eye).
The player then splits the image in two, and then splits and separates channels as needed to make the selected anaglyph version.
It's entirely done in the player.
The Pro: no color compression artifacts messing with the anaglyph.
The Con: half horizontal resolution.

Interestingly, our eyes seem to handle the halved resolution better than the color issues.

If you need anaglyph previews, my suggestion is to hire a programmer to make a player capable of splitting channels in the player. That way, you get rid of the color artifact issue completely. If you have the bandwidth, you could use full res side-by-side or top-bottom config on the video output, which of course would double the filesize, but the image would be full res for each channel.

[Jeff Kilgroe]

That's the way Silverlight does 3D as well -- separate stream for each eye, otherwise you get color anomalies when compressing an anaglyph feed. I really don't know if there's a compression format I would recommend here, but would say the best route would be a custom HTML5 or user-downloadable player plug-in that can assemble the two ocular streams into one image, possibly two full color streams and allow the player to offer over/under, side by side, sequential or anaglyph output all from the same media. That's what Microsoft is doing. They're abandoning Silverlight for HTML5 and will let their HTML5 player do the proper 3D encoding at the client end.

What is your application here? Just web preview of 3D stock selections? Doing a custom player may be viable, but then again, I'm expecting Microsoft's forthcoming solution and others to be available for cheap licensing in the rather near future....

[Phil Bates]

...What is your application here? Just web preview of 3D stock selections?....

Yes, that's it. We only need anaglyph. I'll look into the other options you are suggesting. Thanks Jeff!

Phil

[Rudi Herbert]

Phil,

I do anaglyph to show footage to potential customers who usually buy clips from me based solely on the way the anaglyph looks. I encode H.264 or even Flash and, yes, the colors bleed and the image doesn't look quite that good, but I think most people that are into the business side of 3D understand this and accept/hope/expect the final product will be devoid of such anomalies. On occasions I get customers who ask me for a side by side sample, in which case I just make our corporate logo much bigger to prevent unauthorized use. the thing that's important though is for the anaglyph to convey a proper sense of depth planes and separation, i.e:. recognizable areas of negative and positive parallax, with parts of the image either protruding from or receding into the screen in layman terms. I have to say I don't get that sense in the frame you posted, everything seems to be beyond the screen plane, without any EASILY discernible negative parallax, which does take away from the feel of it. Also, and for what is worth, both green/magenta and blue/yellow anaglyph modes do a far better job at preserving color integrity and decreasing ghosting than red/cyan does, thus increasing the perceived quality of the already compromised compressed file. And let's face it, people who are into 3D enough to buy stock footage or commission assignments, the source of my 3D business, should be enough in the know how to have all types of glasses with them. But I have to say that until Microsoft or other models allow for better accuracy in exhibiting 3D footage online, good, clean, properly separated anaglyph should be enough to carry us, as it is, and will be for the foreseeable future, the only way to view 3D in regular 2D displays.

[Phil Bates]

Thank you Rudi,

I agree with what you are saying and think we'll live with H.264 for now.


Phil,

...the thing that's important though is for the anaglyph to convey a proper sense of depth planes and separation, i.e:. recognizable areas of negative and positive parallax, with parts of the image either protruding from or receding into the screen in layman terms. I have to say I don't get that sense in the frame you posted, everything seems to be beyond the screen plane, without any EASILY discernible negative parallax, which does take away from the feel of it...

I would like to understand this better. So, are you saying you don't easily see the layers of positive parallax depth? I realize there is no negative parallax. Except for the logo, there is no easy way for this scene to provide any. The bushes on the lower left hand corner cannot come forward any further without an uncomfortable break in the screen plane.

Going back to the problem of seeing depth beyond the screen plane, could this be the nature of the subject matter, or perhaps because we have designed the clip with only 1.5% positive parallax (so it can be displayed on larger screens)?

It seems that by the nature of the beast landscapes will live mostly in positive parallax space. I just never considered this a problem but would like to here more.

Thanks again Rudi for looking at this.
Phil

[Rudi Herbert]

Phil,

Well, there's your answer, 1.5% positive parallax on something already far from the camera will give you a diminished sense of depth. This is a conundrum indeed, which medium to shoot for and what amounts of negative parallax, if any, to include in the frame, knowing that depending on the screen size, this negative parallax can either be acceptable or a deal breaker. And right again, most landscapes will live exclusively in the positive parallax domain, but I would suggest trying to fit in a foreground object, such as an isolated bush, flower bud, something, that can converged in front of the screen for smaller screen sizes or dealt with, including rotoscopoed out if needed, when going to cinema sizes. But no, I cannot easily see the different receding layers of depth in your frame, no doubt created by the compression, which is why is all the more important to include foreground elements, even if the foreground could be relatively distant such as in this case, to give a strong depth perception through negative parallax. In hindsight, given your subject matter and intended final medium, I don't think there's much you could have done differently, but yes, the image looks a little flat. Hey, who said 3D was supposed to be fun?

[Rudi Herbert]

...and by the way Phil, I sent you a PM, have a look at it and let me know.