[PYTHON] Create an original Random Dot Stereogram (RDS) from 2D photos with machine learning depth estimation

We have been trying and error in search of the best random dot stereogram (RDS).

-Create an original random dot stereogram (RDS) in Python. -In search of the best random dot stereogram (RDS).

The day has finally come to use machine learning. You should now have the ideal Random Dot Stereogram (RDS).

With the recent boom in machine learning, I'd like to explain what a random dot stereogram (RDS) is for those unlucky who have been researching machine learning and arrived at this article.

What is RDS?

Random dot stereogram (RDS) is a pattern that allows stereoscopic vision by moving the focus of the eye away from or closer to the screen. Everyone should have seen it once. The above figure is good. If you shift the focus of your eyes, the characters Hello, World! Should appear. It is easier to understand if you see it from your PC. By the way, I can see it on my smartphone: v:

I want to generate this random dot stereogram (RDS) from an original 2D photo.

Depth estimation

A depth map is needed to create a random dot stereogram (RDS). Apparently, machine learning technology can be used to estimate depth from 2D photographs. Now you can make the best random dot stereogram (RDS).

Tensorflow When I was researching depth estimation, I found a blog of a person who made a similar attempt. This seems to be done with normal stereograms instead of random dots.

-I tried to create a stereoscopic image from a 2D photo --Memento

Since it is a big deal, I will refer to this method for the depth estimation part. Thank you very much.

I tried to estimate the depth

I estimated the depth of this image.

Here are the results.

Apparently it worked (?!).

Make it RDS

I tried it.

I don't understand the meaning at all. It's a failure.

end

Probably it failed because there was not much shading in the depth map. difficult…