2D to 3D

[Sareesh Sudhakaran]

Sorry if this is a naive question but I'm in the process of doing 2D to stereoscopy conversion tests and I wanted advice on the best method possible on a home PC. My intended purpose is not youtube or a home viewing environment, but a cinema screen.

After some research, I've found some ideas:
1. Change the perspective on the image frame and get two L and R versions.
2. Use the displace filter in photoshop/after effects to get L and R
3. Rotoscope out the foreground painstakingly for each frame (with Mocha in After effects?) and then create the L and R images. But how? Couldn't find a tutorial online.
4. Use avisynth and virtualdub. But I have no idea where to start or whether it can be accurate because each shot is different, right?

Even though I have tested a few of the above with anaglyph at home, it's impossible to judge the best option for screen. Where would a good place to start be? Tried google and the above is all I got. Thanks.

[Adam Stanislav]

Originally Posted by Sareesh Sudhakaran

I wanted advice on the best method possible on a home PC.

Reshoot it all, this time with two cameras and compose it for 3D. Trying to convert 2D to 3D is like trying to convert mono sound to stereo, or B&W to color. The information is not there, all you can do is guess.

More importantly, shooting for 3D is different from shooting for 2D. Not just by using two cameras but composing everything differently. If it was shot in 2D, it should stay in 2D. If it should be in 3D, it should be shot in 3D. Otherwise it is like trying to convert a painting into a statue.

[Steve LaPierre]

While I like Adam's comparison of converting a painting into a statue, Sareesh is probably thinking along the lines of how a TV can convert 2D into "3D" on the fly. There are cues in a conventional image for creating "3D" but I have no idea what program exists to do this and the result would certainly not compare to shooting in stereo in the first place as Adam suggested. It would be interesting to know how companies do the process with their TVs however, maybe someone can explain it. And I agree with Adam that it is pointless for a cinema release as Sareesh has planned.

[Sareesh Sudhakaran]

Reshoot it all

Adam, I understand your point and agree with it completely. Yet, if someone is paying you money to do the impossible, then it can be done in my opinion. I have seen some real interesting results of films converted to 3D as tests. They might not win the respect of stereo-nerds, but they will surely impress the movie-going audience. If someone can find a way to get it done, then he can write his own ticket.

It would be interesting to know how companies do the process

Steve, this is supposedly a trade secret. No post facility will give out this detail, and they are all probably breaking their heads over it. Having seen Clash of the Titans in 3D, I know even Hollywood has no clue when it comes to 2D to 3D conversion. There are software that claim to be able to do it but I doubt their usefulness.

[Tim Dashwood]

I have done a lot of research into 2D to 3D conversion because every once in a while we receive shots that just cannot be saved. I loath it but it is sometimes a necessary evil.

There are many different methods out there:

• High Quality yet tedious wire-frame recreation and texture mapping used in the first 7 minutes of Harry Potter 6 in IMAX3D and Star Wars conversions (I've seen the first 5 minutes of Star Wars IV and it is AWESOME!)
• Medium Quality but faster using rotoscoping and flat 2.5D layers in 3D space used in Clash of the Titans, Alice in Wonderland, Pirhanna 3D and Harry Potter 7 (scrapped last-minute by WB because it was so bad.)
• Moderate to Low quality using depth maps (or luminance map) and pixel shifting (I wrote a plugin to do this in AE and FCP.)
• Low Quality 3D "gimmicky" but realtime conversions using "2D to 3D software" on PC or built into 3DTVs and processor boxes. Every system uses a different type of special sauce but most seem to just work by warping the images to create a planer shift near the centre of the X axis and tricking your brain into thinking the center is closer than the edges.
• An experimental method for changing the angle and shifting a duplicate of one image to appear to be in perfect parallel positive parallax with itself, yet beyond the binocular threshold. This causes the viewer's brain to assume the objects are too far away to have useful parallax points and monocular depth cues automatically kick in creating a 3D depth perception. We are experimenting with this technique and will hopefully have some positive results to demonstrate at NAB.

[Sareesh Sudhakaran]

Tim, I would have bought the stereo3D plug-in a long time ago except it only works on macs.

I think a combination of roto and depth maps should be quite sufficient. But it's tedious.

[Ian Knowles]

Originally Posted by Tim Dashwood

I have done a lot of research into 2D to 3D conversion because every once in a while we receive shots that just cannot be saved. I loath it but it is sometimes a necessary evil.

There are many different methods out there:

• High Quality yet tedious wire-frame recreation and texture mapping used in the first 7 minutes of Harry Potter 6 in IMAX3D and Star Wars conversions (I've seen the first 5 minutes of Star Wars IV and it is AWESOME!)
• Medium Quality but faster using rotoscoping and flat 2.5D layers in 3D space used in Clash of the Titans, Alice in Wonderland, Pirhanna 3D and Harry Potter 7 (scrapped last-minute by WB because it was so bad.)
• Moderate to Low quality using depth maps (or luminance map) and pixel shifting (I wrote a plugin to do this in AE and FCP.)
• Low Quality 3D "gimmicky" but realtime conversions using "2D to 3D software" on PC or built into 3DTVs and processor boxes. Every system uses a different type of special sauce but most seem to just work by warping the images to create a planer shift near the centre of the X axis and tricking your brain into thinking the center is closer than the edges.
• An experimental method for changing the angle and shifting a duplicate of one image to appear to be in perfect parallel positive parallax with itself, yet beyond the binocular threshold. This causes the viewer's brain to assume the objects are too far away to have useful parallax points and monocular depth cues automatically kick in creating a 3D depth perception. We are experimenting with this technique and will hopefully have some positive results to demonstrate at NAB.

I have been roughly modelling scenes (static fixed camera scened mostly) to create depth map and rotoscoping the foreground non static objects. I have also found the if you create a roto solid and use gradient tools on the roto you can dramaticaly reduce the flat looking 3D and if you use the gradient of the floor cieling or nearby object you can adjust your rotoscoped solids brightness to position your object properly. I have had quite good success with that and all though tedious as hell it works quite well. Anything that you can do natively though you should do!

@Tim Dashwood - Care to give a little more detail on your experimental method always open to new ideas