ConceptPose: In-the-Wild 6D Pose Estimation

Upload two images of the same object from different viewpoints and get the estimated relative 6DoF pose.

Pipeline: SAM3 (segmentation) → DepthAnything3 (depth) → ConceptPose (semantic 3D registration)

Anchor Image (reference view)

Query Image (target view)

Object Category

Pre-cached categories: ape, banana, basket, benchvise, bike, book, bottle, bowl, box, cam, camera, can, car, cat, clamp, cracker box, croc, cup, cutlery, driller...

Result

Concept Saliency Visualization

Pose Projection Visualization

Example Pairs

Anchor Image (reference view)	Query Image (target view)	Object Category