A Partial-to-Partial Point Cloud Registration Method Based on Geometric Attention Network

<div>Example results of GAP-Net trained on ModelNet40 and applied to Stanford data. The objects (a)–(d): bunny, hand, happy Buddha, and horse; on the left is the initial position of the two point clouds, on the right is the registration result, and the rotation and translation error of the registration are calculated. (a) RRE = 3.263, RTE = 0.0075; (b) RRE = 0.282, RTE = 0.0183; (c) RRE = 2.232, RTE = 0.0057; (d) RRE = 0.929, RTE = 0.0012.</div>

Journal of Sensors

fig9

Figure 9

Figure 9: A Partial-to-Partial Point Cloud Registration Method Based on Geometric Attention Network 