What is the best depth map generation algorithm?

I am involved in a 2D-3D application project, and I am looking for a way to create a depth map of a single input image without additional information. I know that there is a kind of โ€œartificial intelligenceโ€, but perhaps an efficient algorithm.

At the moment, I have found this: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.109.7959&rep=rep1&type=pdf , but I wonder if there is a better method before embarking on the implementation. Suggestions? Thanks!

+6
source share
2 answers

I have written quite a few automatic depth map generators. I do not think that in all cases is better than everyone else. It all depends on the stereo pair you start with. I personally think that a depth map generator based on a local method (based on a window or block) with smoother edge preservation is probably the best universal depth map generator.

Anyway, on this page:

depth map software

you can find depth map generator software based on optical flow based on weight windows, graphs and many other things that are related to creating a depth map and creating a lenticular shape. The best part is that all this is free.

To convert 2d-3d (which is more than what you are asking for), there is a piece of software called DMAG4 that uses a sparsely populated depth map (usually made in Gimp with a brush) to indicate the main depths and then fill in the blank areas. using interpolation, keeping edges of objects (keeping edges).

DMAG4 can be found here (it's free to use):

Software to convert 2d to 3d DMAG4

Another way to convert 2d to 3d is to use a program like Gimpel3d or Blender as a free one. Clearly, this is beyond the scope of the depth map, since you are essentially creating a 3D scene in which you can move around (using camera movement in Blender). This is often called "camera mapping."

+4
source

Ok, I recently came across this:

http://make3d.cs.cornell.edu/code.html

which comes with the code, although the license may be too restrictive ("Nonprofit - you cannot use this work for commercial purposes").

the gallery is impressive.

http://make3d.stanford.edu/images/showall

+1
source

Source: https://habr.com/ru/post/900371/


All Articles