search

MobileRoboticsSkoltech / map-metrics

Pip-package for trajectory benchmarking from "Be your own Benchmark: No-Reference Trajectory Metric on Registered Point Clouds", ECMR'21
Other
37 stars 6 forks source link

Speed up mme and mpv #1

Open anastasiia-kornilova opened 3 years ago

anastasiia-kornilova commented 3 years ago

Now, the most time consuming operation is iteration over points in point cloud, which could be parallelized.

Razlaw commented 3 years ago

Hi, an already parallelized version can be found here ;) It's C++ though and the MPV is implemented a little differently, but the description in the paper isn't very detailed. So that's totally the authors' fault :) Let me know if you need help deciphering the code, it's pretty old.

anastasiia-kornilova commented 3 years ago

@Razlaw , hello! Thanks for your comment! Actually, I keep in mind this code as good efficient implementation from the moment I started doing this research project :)

The current stage of the work is a proof-of-concept, the next steps will be more about engineering and creating a tool. I see this as a tool with a Python interface for possible application in ML/DL as a metric, so it should be fast and, maybe, parallelized on GPU also. So, we will be happy to reuse your code and provide bindings to it at least for CPU part.