Strength
What it is useful for
- Requires no exact point correspondences and suits generated or completed shapes.
- Penalizes both missing and extraneous regions.
Shape reconstruction / auxiliary metric · shape
Bidirectional Chamfer Distance
Uses bidirectional nearest-point matching between predicted and ground-truth shape point clouds to measure reconstruction similarity.
Primarily evaluates NOCS, shape completion, or category-level shape reconstruction rather than 6DoF pose alone; read it alongside pose metrics.
Principle
The formulas define the computation. The diagrams show whether the error concerns a 3D model, visible surfaces, the image plane, or aggregation across thresholds.
Compute nearest-neighbor distances from prediction to ground truth and from ground truth to prediction, ensuring both sides are covered.
Can reveal missing regions or local displacements, but does not guarantee pointwise topological correspondence.
How to compute
Place both shapes in the same NOCS or world coordinate frame, and sample according to the protocol.
Compute nearest-neighbor distances in both directions.
Sum or average to obtain CD.
Strength
Caveat
Paper atlas
Links are derived by normalizing the metric names used in each paper. Directly reported metrics are distinguished from metrics associated through a broader protocol.
ICRA · Ethan K. Gordon, Bruke Baraki, Hien Bui et al.
ECCV · Li Jin, Yuchen Yang, Weikai Chen et al.
3DV · Hiranya Garbha Kumar, Minhas Kamal, Balakrishnan Prabhakaran
3DV · Ayce Idil Aytekin, Helge Rhodin, Rishabh Dabral et al.
3DV · Licheng Shen, Saining Zhang, Honghan Li et al.
3DV · Jane Wu, Georgios Pavlakos, Georgia Gkioxari et al.
ICRA / IEEE RA-L · Yibo Liu, Zhaodong Jiang, Binbin Xu et al.
3DV · Weikun Peng, Jun Lv, Cewu Lu et al.
IJCV · Jian Liu, Wei Sun, Hui Yang et al.
ECCV · Tzu-Yuan Lin, Ho Jae Lee, Kevin Doherty et al.
3DV · Suhas Gopal, Rishabh Dabral, Vladislav Golyanik et al.
CVPR · Jingnan Shi, Rajat Talak, Harry Zhang et al.
3DV · Chin-Hsuan Wu, Yen-Chun Chen, Bolivar Solarte et al.
3DV · Xianghui Xie, Jan Eric Lenssen, Gerard Pons-Moll
ICRA · Katherine Liu, Sergey Zakharov, Dian Chen et al.
IROS · Marios Glytsos, Panagiotis P. Filntisis, George Retsinas et al.
ICCV · Dimitrije Antić, Georgios Paschalidis, Shashank Tripathi et al.
3DV · Nathaniel Chodosh, Anish Madan, Simon Lucey et al.
ECCV · Chuanrui Zhang, Yonggen Ling, Minglei Lu et al.
CVPR · Yijia Weng, Bowen Wen, Jonathan Tremblay et al.
3DV · Stefan Ainetter, Sinisa Stekovic, Friedrich Fraundorfer et al.
CVPR · Jingtao Sun, Yaonan Wang, Mingtao Feng et al.
CVPR · Liyuan Zhu, Shengyu Huang, Konrad Schindler et al.
IROS · Francesco Milano, Jen Jen Chung, Hermann Blum et al.
Next
Compare metrics at the same level to understand why one score improves while another stays unchanged.
3D Bounding-box Intersection over Union
Average Distance of Model Points
Average Closest Point Distance