3d cad object만 존재하는 상황에서 oneshot 6d pose estimation을 어떻게 할것인가?
이 그림 하나면 모두 설명 가능하다.
Stage 1. 전형적인 Few shot segmentation 방법 사용했다(Hypercorrelation Squeeze for Few-Shot Segmentation참고).
Stage 2. Templates for 3D Object Pose Estimation Revisited: Generalization to New Objects and Robustness to Occlusions 논문과 거의 완전 동일한 방법 사용했음, template matching
Stage 3. RNNPose: Recurrent 6-DoF Object Pose Refinement with Robust Correspondence Field Estimation and Pose Optimization 논문과 비슷한 접근이다. Img상에서 2d-2d corrsespondence field 만들어서 접근함
Reference
1. A Multi-Stage One Shot Object Pose Estimation Framework
2. Templates for 3D Object Pose Estimation Revisited: Generalization to New Objects and Robustness to Occlusions
3. Hypercorrelation Squeeze for Few-Shot Segmentation
4. RNNPose: Recurrent 6-DoF Object Pose Refinement with Robust Correspondence Field Estimation and Pose Optimization
