Images and camera poses as input
somepablo opened this issue · comments
Hi, first of all congrats for this amazing library.
I've seen that you've made this lib as an extension to MVE, so you take as input the result of sfmrecon (point cloud, matches, camera poses...).
What about if I already have images and known camera poses, would it be easy to adapt this lib for that?
Many thanks
You can just create an MVE scene manually and add the camera poses and images. Here's the documentation for the format: https://github.com/simonfuhrmann/mve/wiki/MVE-File-Format
So, aren't sparse points or covisibility indexes required for smvs? Just images and camera matrixs are enough?
Yes, images and camera matrices are enough. SMVS runs a coarse semi-global matching with spatially close cameras if no other visibility information is found. But you have to be a little careful, this requires a reasonable approximation of the depth range in the scene which you can set with --sgm-range=min,max
.
Edit: If this doesn't work well, you can also try to use the MVE application featurerecon
which will generate a sparse point cloud and visibility information for known camera parameters.
Well what I'm planning is to adapt this part of the 3d reconstruction pipeline to some real-time SLAM algorithm.
I want to record frames and camera poses during a slam execution in a mobile device and later, send that info to a server that executes SMVS without the first MVE part as I already know the structure and motion.
Do you think this would be possible with this lib?
I know I can just save frames and run the complete MVE pipeline, but I want to keep my original SLAM camera positions and scale.
Yes, this should work.
Thank you,
What about if camera poses provided by the SLAM algorithm are a little bit drifted? Will SMVS try to correct that by aligning frames or something like that?
I'll try to reconstruct small rooms mainly (about 15m2). What sgm-range do you think I should use?
No SMVS cannot correct for errors in the camera poses.
The sgm-range depends on the actual scale of your scene (the scale in which the camera poses are saved). If you can scale it to, for example, real meters you can just select something like the max dimension of your room, maybe around 0.1-8m.
Closing due to inactivity.