How to choose between 3D Object Pose GEMS?

Hello,
Could someone explain me the differences between :

Which is the fastest ? Which is the more accurate ?

Thank you

3D Object Pose Estimation with Pose CNN Decoder and 3D Object Pose Estimation with AutoEncoder both take an input RGB image and gives out the estimated 3D pose of an object using Augmented Autoencoders. And 3D Object Pose Estimation with Pose CNN Decoder is the most recent one and is the most accurate.

3D Object Pose Refinement is used to refine and improve the accuracy of the estimated 3D pose from above by using depth image. It takes RGB image, depth image as well as initial pose estimate as input and uses an iterative search to refine the pose further using the depth information. So you can use this as second step after pose estimation to improve accuracy of the estimated pose or just do the step 1 by estimating the 3D pose form single RGB image.

1 Like

Thank you ! do you have an idea on how many fps can I get running the Pose estimation and refinement with Jetson nano ?