FocalPose++ is a novel method that accurately estimates both camera focal length and object pose from a single RGB image by extending the render-and-compare approach with focal length update rules, a disentangled training loss, and a synthetic data generation strategy based on real data distributions.
The proposed TransPose framework exploits Transformer Encoder with a geometry-aware module to develop better learning of point cloud feature representations for accurate 6D object pose estimation.
Proposing SD-Net for accurate 6D pose estimation in bin-picking scenarios through symmetric-aware keypoint prediction and domain adaptation.