How does dm-vio pick points to track?

[GilMe]

Hello,
I was going over both the code and the paper and couldn't figure out how dm-vio picks the points it uses to track in the various frames. Does it use any kind of feature detector to correlate between points in images? What kind of feature detector does it use?

Thanks for the response

[lukasvst]

DM-VIO uses the same point selection and tracing as DSO, so you can checkout the details here under 3.2 point management: https://arxiv.org/pdf/1607.02565.pdf

In short: As it is a direct method it can track any points with sufficient gradients and no feature detector needs to be used. If you want it to track specific features you can modify it however, similar to how it is done in LDSO.