NVlabs / FoundationPose

[CVPR 2024 Highlight] FoundationPose: Unified 6D Pose Estimation and Tracking of Novel Objects
https://nvlabs.github.io/FoundationPose/
Other
955 stars 99 forks source link

Hello and thank you for your contribution.My question is, how do you distinguish between objects that look the same but are different in size?(cad model available) #95

Closed wsq1010 closed 1 week ago

wenbowen123 commented 2 weeks ago

you can run estimation once using each of the CAD models (Say there are N models, you run totally N times for each of them, and you get N best poses), then send their poses together to the score network to choose. The highest score among the N poses corresponds to the best CAD model.