Reliable Inlier Evaluation for Unsupervised Point Cloud Registration

Abstract

Unsupervised point cloud registration algorithm usually suffers from the unsatisfied registration precision in the partially overlapping problem due to the lack of effective inlier evaluation. In this paper, we propose a neighborhood consensus based reliable inlier evaluation method for robust unsupervised point cloud registration. It is expected to capture the discriminative geometric difference between the source neighborhood and the corresponding pseudo target neighborhood for effective inlier distinction. Specifically, our model consists of a matching map refinement module and an inlier evaluation module. In our matching map refinement module, we improve the point-wise matching map estimation by integrating the matching scores of neighbors into it. The aggregated neighborhood information potentially facilitates the discriminative map construction so that high-quality correspondences can be provided for generating the pseudo target point cloud. Based on the observation that the outlier has the significant structure-wise difference between its source neighborhood and corresponding pseudo target neighborhood while this difference for inlier is small, the inlier evaluation module exploits this difference to score the inlier confidence for each estimated correspondence. In particular, we construct an effective graph representation for capturing this geometric difference between the neighborhoods. Finally, with the learned correspondences and the corresponding inlier confidence, we use the weighted SVD algorithm for transformation estimation.Under the unsupervised setting, we exploit the Huber function based global alignment loss, the local neighborhood consensus loss and spatial consistency loss for model optimization. The experimental results on extensive datasets demonstrate that our unsupervised point cloud registration method can yield comparable performance.

Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration

“…To handle ground-truth labeling issues, great efforts [12,21,23,39,45,49] have been devoted to unsupervised deep point cloud registration. The existing methods mainly lie in autoencoders [12,21,39] with a reconstruction loss or contrastive learning [10,14,47] with data augmentation. Although encouraging results have been achieved, some limitations remain to be addressed.…”

“…Baselines. We chose recent supervised SOTA methods: DCP-v2 [44], OMNet [48], RPM-Net [50], Predator [19], REGTR [51], CoFiNet [52], and GeoTransformer [38], as well as unsupervised method RIENet [39] and UGMM [20] as our baselines. For traditional methods, we choose pointlevel methods ICP [5] and FGR [57], as well as probabilistic methods CPD [34], GMMReg [22], SVR [7], and FilterReg [18] as baselines.…”

“…In REGTR, Predator samples 450 points in the experiment, and OMNet obtained a slightly improved result in all cat- 2 reports registration results on ModelNet40, in which the best results for each criterion are labeled in bold, and the best results by unsupervised methods are underlined. We compare against the recent unsupervised [39] and supervised [19,38,44,48,51,52] methods. When compared with unsupervised methods, our UDPReg outperforms the correspondence-based CEMNet, RIENet and GMM-based UGMM [20] in all metrics under both normal overlap (ModelNet) and low overlap (ModelLoNet) regimes.…”

“…Moreover, the success of learningbased methods mainly depends on large amounts of ground truth transformations or correspondences as the supervision signal for model training. Needless to say, the required ground truth is typically difficult or costly to acquire, thus hampering their application in the real world [39].…”

HTMC: hierarchical tolerance mask correspondence for human body point cloud registration

“…Predator ( Huang et al, 2021 ) and GeoTansformer ( Qin et al, 2022 ) search for correspondence pairs from coarse to fine. RIENet ( Shen et al, 2022 ) searches for reliable correspondence pairs. These works handle point cloud registration in various aspects.…”