Convert 2D


With the popularity of movies converted from 2D to 3D these days (Men In Black III, The Avengers, and many more), I figured I’d embrace the technology rather than fight it. I still definitely believe there’s no way to capture all of the realistic levels of detail in a 2D conversion that you could’ve gotten by just shooting it in 3D in the first place… of course that’s not possible with old material, there is a huge demand to convert the 2D classics, and it can be well done!


1st conversion - hallway 2D Original
I chose this seemingly simple 2D shot for my first-ever attempt at doing a 2D to 3D conversion, but the fact that it was taken closer to the ceiling than the floor (at standing human height) led to creating a more complicated depth map. If the photo was taken from the XY center, I could have simply used a single square gradient tool spun to 45ยบ & stretched to match dimensions. Next is the solution I came up with…


1st conversion - hallway Depth Map
Here is the depth map I created from the original 2D hallway image. A bit of geometry was involved in intersecting 4 different gradient layers & finding the correct ‘blending mode’, but I suppose this is still quicker & neater than trying to paint a depth map by hand, as any variations in levels would show up.


1st conversion - hallway 3D Version
This final anaglyph was created by using the ‘Filter>Distort>Displace’ feature in Photoshop to offset the 2D image in amounts according to pixel brightness levels in the previous Depth Map. You actually only need to synthesize one additional image in this manner (using the original as the other (left or right)), but for this example I pretended the original is the ‘center’ image, and created both left and right eye views from it by displacing a positive amount in one, and a negative amount in the other.


Since this & other early successes in conversion, I’ve learned about an entirely different method of converting 2D into 3D, and that involves a significant amount of time & detail work. Basically it involves ‘slicing’ your image with concentric selections into many, many layers from front to back, and shifting each one to the side by a very small amount. Then you combine this new image with the original for a stereo pair. The results I’ve seen are promising indeed, as this could be the method of choice for faces, irregular details, etc.