OPEN CALL: Commissioned Optical Flow 3D / Scene Flow Implementation


#1

Hey VVVVriends,

here comes the short-version of the purpose of this thread:

A - Find a gifted coder who is willing and confident to do implementation of a Optical Flow 3D / Scene Flow implementation for VVVV during the coming 4 weeks.

B - Use the power of crowdfunding as demonstrated lately HERE, to collect enough budget to financiate the implemention and to release the result as a contribution for everyone in the community.

And here comes the full description:

We are currently working on an installation-idea for a museum including this. And since it is supposed to be an interactive piece, a combination of the fluid-sim with a RGB-D-camera (f.e. Kinect) and a 3D Optical Flow is - well - kind of mandatory.

We did some research, collected some papers and examples and spoke with @everyoneishappy and @otosatram and @tmp about the topic. And since them and us are to busy at the moment to deep-dive into the wonderful world of Calculus, Derivatives and co. we thought: Maybe some fit but bored coder here in the community is interested to do a proper and funded implementation.

And because there are plenty of other usecases for a scene flow implementation, we are assuming, that some community-members are maybe not capable of coding it but super interested to having a Scene Flow-module in VVVV and like to co-fund the project.

So this the plan: At first we have to find who could do it. Then she or he has to give a quote, what does it take to do it. And then we see who else - besides us - would support this contribution.

So, either you are A or B, please leave a comment or get in touch which us and feel free to post you ideas about the topic or a suggestion of someone who could do it.

Last but not least some of the papers/repos/links we’ve gathered during the research:

RGB-D Flow: Dense 3-D Motion Estimation Using Color and Depth (looks very promising to me)
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.308.9908&rep=rep1&type=pdf

–> Code of this paper a little bit down here:
https://homes.cs.washington.edu/~xren/

Overview of interesting papers, not only scene flow:
http://vision.in.tum.de/research/rgb-d_sensors_kinect
http://vision.in.tum.de/research/sceneflow

–> Including:


and the code-> https://github.com/MarianoJT88/PD-Flow

A 6 year old test, but looks not to bad though.

Dense Semi-Rigid Scene Flow Estimation from RGBD images (results looks good):

Scene Flow Estimation: A Survey


#2

Same here! super interesting!


#3

First test looks promising:

:)


#4

Thanks to Mariano Jaimez from TU Munich, who gave me one major hint how to tackle the velocity for Z, the implementation was pretty straight forward and at the end just few lines of code.

The whole thing will be part of a contribution, including also the Fluid3D.

Here a first recent test in 3D together with the fluid-sim:

So, thread closed and thanks for reading.


#5

wow awesome. I was reading the papers and about to propose that I can do it in in autumn, but that seems not neccesary anymore :)


#6

Has this contribution been completed yet?


#7

oh my god that is so much win!


#8

Just in case you missed it: https://vvvv.org/contribution/opticalflow3d