… We explore the robustness of PointNetLK against Gaussian noise on points. The experiment
set-up is as follows: a template point cloud is randomly sampled from the faces of the …

… Robustness to noise: To verify our method’s robustness to noise, we trained it on noiseless
… Compared with the original PointNetLK and DeepGMR, our approach is more robust to noise …

… A visualization of point cloud registration using (a) PointNetLK Revisited [10] and (b) PE on
ModelNet40 dataset. Gaussian noise with a standard deviation of 0.05 is presented. Green: …

… For PointNetLK, we adopt similar procedure suggested by the authors to sample visible
points in the other point cloud. However, we sample visible points from the points with a …

… [1]: To compare with the registration method of using deep learning, we use PointnetLK,
which combines Pointnet with Lucas-Kanade algorithm. We train the PointnetLK using the same …

… rotation error as compared to ICP and PointNetLK, which shows that our iterative PCRNet
is robust to Gaussian noise. It is worth noting that PointNetLK performs the worst and is very …

… PointNetLK directly optimizes the distance of aggregated features … However, we think that
PointNetLK and DirectNet do not … Lucey, “Pointnetlk: Robust & efficient point cloud registration …

… 8, we show that retrofitting PointNetLK with MaskNet improves its performance sig… for
PointNetLK compared to PointNetLK when retrofitted with MaskNet (referred to as Mask-PointNetLK)…

… If the large-scale scenes have sparse features, our voxelized analytical PointNetLK will
likely not capture robust features across the entire 3D space, thus leading to bad registrations. …

… We observe that PointNetLK is just another special case of our framework. If we were to use
an … This shows that i-PCRNet is more robust to noise as compared to PointNetLK. Fig. 9(c) …