thanks! Still a lot of refinement left to do (this was initially released as a proof of concept). But I'm happy to see people from all over enjoying the gimmick : )
What type of data did you have to collect? (photo + actual position/orientation of the tube) ; how does this generalize to items of different shapes/sizes? (I think that's something that would be good to show in your demo video)
I hope to do a technical writeup of my process soon. In short, the training images were created from videos of rotating the tube around in hand at various angles and lighting conditions. These images were then annotated to determine the location of several keypoints around the paper. It generalizes pretty well, see here: https://twitter.com/2020cv_inc/status/945502116335906816
7 comments
[ 0.24 ms ] story [ 28.0 ms ] threadIf on Android, just hide inside your Tauntaun and pray... joking (we hope to get this running on Android soon too)