search

Hitachi-Automotive-And-Industry-Lab / semantic-segmentation-editor

Web labeling tool for bitmap images and point clouds
MIT License
1.77k stars 433 forks source link

big point clouds/performance issue #174

Open nullkatar opened 2 years ago

nullkatar commented 2 years ago

Hello @dmandrioli!

Thanks a lot for you great job, the tool is really easy to use and very useful! I was trying to annotate really big point clouds which include ~10 million points and the tool was lagging dramatically while CPU and GPU utilization was really small. I want to ask if I can somehow get rid of this lags probably by configuring something in source code or anything? Smaller clouds is the last option which I would like to consider :(

weimzh commented 2 years ago

I haven't tried 10 millions but I have made the following optimizations for point clouds with ~1 million points, so far working well (still a bit laggy on some lower spec computers but toleratable):

-rewritten point projecting with webassembly (https://gist.github.com/weimzh/95f3ef2f5f57dd94665e8bc81b3e32d2) -replaced ES6 Map/Sets with bitmaps (https://gist.github.com/weimzh/d972c188a42f66816559c427881639e0) or arrays -maybe others misc optimizations as well.

The results are also not saved int ascii pcd file but our internal binary format (again with webassembly).