Zero-Shot Object Detection for 6D Pose Estimation

[eliastreis]

Problem We're integrating FoundationPose for 6D object pose estimation, which needs a bounding box for the target object. Since we're working with an open set of objects, we want to avoid model re-training/refinement and are looking for a zero-shot object detection model. We've tried foundational models like DINO, SAM, and their Grounded versions but are still facing issues with reliability and robustness, requiring manual verification.

Question Do you have experience with this / is there a zero-shot detector or related approach which works for you?

[abhishekmonogram]

@eliastreis I have tried CNOS and works decently well with the data I have been using FoundationPose for. Check if that works for you