TABLE III Results obtained for the testing point clouds from the partial models on ModelNet40 under all categories [44]. METHOD PointNetLK [11] RATE 1% 5% 10% 50% 80% PCRNet [30] 1% 5% 10% 50% 80% CFNet(ours) 1% 5% 10% 50% 80% RMSE(R) 10.652 12.045 11.807 12.181 10.603 12.356 10.486 8.486 5.616 4.437 5.613 3.846 2.453 1.501 1.143 MAE(R) 3.152 3.258 3.172 2.657 2.472 10.617 8.175 6.834 4.592 3.865 4.373 3.812 1.698 1.076 0.934 RMSE(t) 0.3468 0.4256 0.3894 0.3887 0.3885 0.0486 0.0274 0.0295 0.0274 0.0527 0.0128 0.0102 0.0086 0.0042 0.0036 MAE(t) 0.3286 0.3967 0.3964 0.3823 0.3841 0.0386 0.0371 0.0191 0.0337 0.0482 0.0141 0.0092 0.0074 0.0038 0.0034 translation branch under this condition. Even so, our network is far ahead of most algorithms in terms ofthe rotation error metrics.