Point cloud completion, aiming to reconstruct missing part from incomplete point clouds, is a pivotal task in 3D computer vision. Traditional supervised approaches often necessitate complete point clouds for training supervision, which are not readily accessible in real-world applications. Recent studies have attempted to mitigate this dependency by employing self-supervise mechanisms. However, these approaches frequently yield suboptimal results due to the absence of complete structure in the point cloud data during training. To address these issues, in this paper, we propose an effective framework to complete the point cloud under the guidance of self learned complete structure. A key contribution of our work is the development of a novel self-supervised complete structure reconstruction module, which can learn the complete structure explicitly from incomplete point clouds and thus eliminate the reliance on training data from complete point clouds. Additionally, we introduce a contrastive learning approach at both the clusterand instance-level to extract shape features guided by the complete structure and to capture style features, respectively. This dual-level learning design ensures that the generated point clouds are both shape-completed and detail-preserving. Extensive experiments on both synthetic and real-world datasets demonstrate that our approach significantly outperforms state-of-the-art self-supervised methods.

Point cloud completion is essential for robust 3D perception in safety-critical applications such as robotics and augmented reality. However, existing models perform static inference and rely heavily on inductive biases learned during training, limiting their ability to adapt to novel structural patterns and sensor-induced distortions at test time. To address this limitation, we propose PointMAC, a meta-learned framework for robust test-time adaptation in point cloud completion. It enables sample-specific refinement without requiring additional supervision. Our method optimizes the completion model under two self-supervised auxiliary objectives that simulate structural and sensor-level incompleteness.

Point cloud completion is essential for robust 3D perception in safety-critical applications such as robotics and augmented reality. However, existing models perform static inference and rely heavily on inductive biases learned during training, limiting their ability to adapt to novel structural patterns and sensor-induced distortions at test time. To address this limitation, we propose PointMAC, a meta-learned framework for robust test-time adaptation in point cloud completion. It enables sample-specific refinement without requiring additional supervision. Our method optimizes the completion model under two self-supervised auxiliary objectives that simulate structural and sensor-level incompleteness.

We further introduce a multi-scale diversity penalty that operates in the feature space of our discriminators.

Since all our experiments (e.g., semantic completion, surface reconstruction, global/part-wise manipulation) are evaluated on the PCN dataset [ Similarly, we also test the "Chair" and the "Table" models on chairs and tables extracted from the ScanNet [ PoinTr is shown in Figure 2 and Figure 3. Qualitative comparisons of part segmentation are visualized in Figure 5, which shows that our SPoVT correctly completes each part of the point clouds. Table 1: Evaluation of point number distributions in predicted point clouds. We now provide more qualitative visualization results on point cloud completion, surface reconstruction, and part-wise manipulation in Figure 6, Figure 7, and Figure 8, respectively. On the other hand, the Alpha value can be chosen as 0.01 for our results with Note that the first three columns are chairs and the last three columns are tables.Figure 4: Architecture of our proposed Refiner θ Figure 5: Qualitative results of completed point cloud with predicted part labels. Figure 6: Qualitative evaluation of completed point cloud.

As one of the most commonly used data representations, point clouds can be easily acquired by 3Dsensors.