Termination: FAILURE

[swarth100]

Hello @mihaibujanca ,

We are currently working on a project which will heavily rely on your library to perform dynamic reconstruction of non-solid objects.

By running the sample umbrella test we get two successful Termination: CONVERGENCE, yet straight after we start getting continuous failures. Eventually (~frame 40) it all falls apart with nans everywhere and an assertion failure.

The following is a quick dump of the average log we have: https://pastebin.com/iNjY1ad6

And on another machine: https://pastebin.com/RsDz2gPz

We were wondering if you knew what the issue could be. Thanks!

[mihaibujanca]

Hi Alberto,

Happy to know you'll be using the code :) - out of curiosity, what's your project about?

I'm currently working on a version that is using Opt (https://github.com/niessner/Opt) for optimisation and I'm hoping I'll get it out early next week, so might be worth waiting for that. It should be a lot faster than the current code that's using Ceres.

In the meanwhile, I have some ideas about why that happens but will need to look into it properly.

Mihai

[swarth100]

Thank you for your prompt reply Mihai!

The final aim of our project will be to reconstruct the movements of (non-solid) entities for research purposes.

Over the next two months we will be working heavily on your library (forked at https://github.com/IC-362-11/dynamicfusion), so please let us know if there's anything we could help on!

Alberto

[swarth100]

Dear Mihai,

Just wanted to point out that we still get termination failures on all machines when running the library with example data after having pulled the latest #25