GistNoesis / BrickRegistration

Then I convert the STL to the URDF : python3 generateURDF.py which under the hood use https://pypi.org/project/object2urdf/ (pip install object2urdf )

These two commands take ~30min and 6hours to run. The generateURDF use GPU and crash on some files if there is not enough memory (GTX 1080 Ti is OK)

It successfully generates ~14000 urdf files provided that your machine is beefy enough. I have create a compressed 1.4GB legoSTLandURDF.tar.gz archive with them https://easyupload.io/h27vgz (This link will expire in 30 days) (I'll try to add a torrent link soon)

It will create a neural network, train it, and run it and do some clustering to identify the bricks. It is intended as a boiler plate to experiment with object recognition to get you started :) Only a mock network, on a mock dataset have currently been run so some bugs are probably hiding.

Leave a Comment