aperezlebel/stereomatching

Disparity calculation using Loopy Belief Propagation.

Disparity calculation using Loopy Belief Propagation

Context

The purpose of this work is to compute the disparity between two images (from which can be recovered the depth of field of the objects in the image).

Input

The input are two images of the same scene taken from two different points. Be careful, it is assumed that the two points differ only by an horizontal translation. Here are the two left and right tsukuba images used for the test :

Try with your own images by replacing the imgL.png and imgR.png files in the input folder.

Compute disparity

To compute the disparity map run :

python main.py

Don't forget to adjust your parameters at the beginning of main.py.

Find your disparity map in the output folder. Here is an example of disparity map :

Resource

This work has been made following the assignment.pdf file, proposed by the computer vision course of the ENPC school. See more details in it.

stereomatching's Issues

The message_update function donot have a smoothness term

I have been going through your code, and I believe your message-update scheme for min-sum LBP is slightly wrong for two reasons :

1. The smoothness/Binary Potential term is missing in the message update.
2. For a message of a given label, it should take minimum of all labels. But your code doesnot have any such scheme.

Can you please look into the code and check, because I am using your code for message update in some other application?