View-Graph Optimization

[macp...@gmail.com]

Hello Chris,
thank you for the wonderful library!

I just read your paper "Optimizing the Viewing Graph for Structure-from-Motion". Are you planning to include this functionality in the future? I would be interested if you think that this approach could improve the global SfM modules implemented in Theia at the moment. Do you think that such an optimization would also work in the case of high outlier ratios (meaning a lot of wrong edges in the view graph)?

Lately I've been experimenting with Cui et al.'s paper on global scale estimation, which seems to yield good results for the global pose, but the whole approach is still very dependent on the validity of the epipolar geometries. And my impression is that a simple triplet filter is very straight forward to implement, but has its limitations.

Yours sincerely,
Phil

[Chris Sweeney]

Hi Phil,

I don't have any plans to incorporate that paper into Theia, since it required some significant changes to core parts of the library in order to perform the algorithm so it doesn't fit well with the generality of the library. We found the procedure described in the paper to work well for two reasons: L1 optimization and better calibration (described in the paper). It would definitely improve the global sfm results of the Theia lib if it were incorporated.

Global SfM will always be subject to outliers and noise in the epipolar geometry. Cui's method has a lot of filtering they do to try and ensure that only valid geometries are passed on to global sfm.

Chris

--
You received this message because you are subscribed to the Google Groups "Theia Vision Library" group.
To unsubscribe from this group and stop receiving emails from it, send an email to theia-vision-lib...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[macp...@gmail.com]

Chris, thank you for your fast answer, ok I see. Anyway, your approach in the paper seems very promising!

Kind regards
Phil