Optical flow to get depth map from stereo pair (DMAG)

[@LutzSchafer]

One of the unfortunate general problems with stereo disparity optical flow solutions is when under presence of occlusions.

[@vanillagorilla8696]

Can it be used with video?

[@ugocapeto3d]

no, it's only for a pair of static images.

[@vanillagorilla8696]

Fair enough, I found some other stuff in my studies.

[@importon]

I find I can get better results than DMAG by using RAFT optical flow and changing the colors of FlowVis to be greyscale.

[@ugocapeto3d]

Do you have a link to what you use?

[@importon]

@@ugocapeto3d It keeps deleting my comments when I try to tell you where to find it. I'm not really sure what to do

[@ugocapeto3d]

@@importon oh yeah, i forgot about that. you can't put links in comments. maybe give the keywords to find it in google or send me link (address is on sidebar on my website 3dstereophoto.blogspot.com in about me box). Thanks in advance.

[@ugocapeto3d]

Below is email that importon was kind enough to send me: --------------------------------------------------------------------------------------------------------------------------------------------------------- Here's the link to RAFT o-flow. To get the result into greyscale mess colors in the flow_viz code. github.com/princeton-vl/RAFT Oh, and there's also this, which looks super promising but I was not able to get it running locally. If you managed to figure it out, I'm sure it would make another excellent tutorial video ;) github.com/princeton-vl/raft-stereo ----------------------------------------------------------------------------------------------------------------------------------------------------------

[@valdoraudsepp953]

Multiple scene photo differentiable diffusion seems to give also very reliable depthmaps. GitHub - diffdiffdepth. Tried to get it work on my Linux (Ubuntu, Mint) desktop PC but failed. Seems that deep knowledge of Python is needed.

[@ugocapeto3d]

Here is the link: github.com/brownvc/diffdiffdepth That looks really good. It's multi view though. Would be interested in seeing what you typically get with just 2 views. Hopefully, they will set up a google colab so that the rest of us can try it.