theHamsta
commented
4 years ago Hi was using the algorithm for the movement of contrast agent in C-arm reconstructions. It works more or less for that application since the signal structure is very smooth (not harsh edges for which DualTVL1 would be better). The implemation is the one from OpenCV extended to the 3d case.
The algorithm does polynomial fitting which worked good for my smooth signals. You would have a differnt use-case when you try to track edges.
Optical flow can only estimate motion with pattern within a few pixel ranges and is very prone to errors since the technique is very localized.
To test is you could just create random advections of your 3d volume to test whether it can capture your simulated motion patterns.
Hello Stephan!! (@theHamsta) Thanks for sharing this implementation with the community. I was wondering in which kind of application you used the 3d optical flow estimation
Problem description: I would like to understand/validate the optical flow in 3d medical images. It would be a great contribution because it is relatively unexplored domain.
Question: From my results up to now, I am not yet convinced that it works as it should be. Do you have any ideas how I can validate that the implemented algorithm works ?
Let me know and thanks again!!!!
All the best :)
N.A.