PARC: A Two-Stage Multi-Modal Framework for Point Cloud Completion
| dc.contributor.author | Cai, Yujiao | en_US |
| dc.contributor.author | Su, Yuhao | en_US |
| dc.contributor.editor | Christie, Marc | en_US |
| dc.contributor.editor | Pietroni, Nico | en_US |
| dc.contributor.editor | Wang, Yu-Shuen | en_US |
| dc.date.accessioned | 2025-10-07T05:03:18Z | |
| dc.date.available | 2025-10-07T05:03:18Z | |
| dc.date.issued | 2025 | |
| dc.description.abstract | Point cloud completion is vital for accurate 3D reconstruction, yet real world scans frequently exhibit large structural gaps that compromise recovery. Meanwhile, in 2D vision, VAR (Visual Auto-Regression) has demonstrated that a coarse-to-fine ''nextscale prediction'' can significantly improve generation quality, inference speed, and generalization. Because this coarse-to-fine approach closely aligns with the progressive nature of filling missing geometry in point clouds, we were inspired to develop PARC (Patch-Aware Coarse-to-Fine Refinement Completion), a two-stage multimodal framework specifically designed for handling missing structures. In the pretraining stage, PARC leverages complete point clouds alongside a Patch-Aware Coarse-to- Fine Refinement (PAR) strategy and a Mixture-of-Experts (MoE) architecture to generate high-quality local fragments, thereby improving geometric structure understanding and feature representation quality. During finetuning, the model is adapted to partial scans, further enhancing its resilience to incomplete inputs. To address remaining uncertainties in areas with missing structure, we introduce a dual-branch architecture that incorporates image cues: point cloud and image features are extracted independently and then fused via the MoE with an alignment loss, allowing complementary modalities to guide reconstruction in occluded or missing regions. Experiments conducted on the ShapeNet-ViPC dataset show that PARC has achieved highly competitive performance. Code is available at https://github.com/caiyujiaocyj/PARC. | en_US |
| dc.description.number | 7 | |
| dc.description.sectionheaders | Creating and Processing Point Clouds | |
| dc.description.seriesinformation | Computer Graphics Forum | |
| dc.description.volume | 44 | |
| dc.identifier.doi | 10.1111/cgf.70266 | |
| dc.identifier.issn | 1467-8659 | |
| dc.identifier.pages | 10 pages | |
| dc.identifier.uri | https://doi.org/10.1111/cgf.70266 | |
| dc.identifier.uri | https://diglib.eg.org/handle/10.1111/cgf70266 | |
| dc.publisher | The Eurographics Association and John Wiley & Sons Ltd. | en_US |
| dc.subject | CCS Concepts: Computer vision → Reconstruction; Machine learning → Neural networks; Information systems → Multimedia information systems; Multimedia and multimodal retrieval | |
| dc.subject | Computer vision → Reconstruction | |
| dc.subject | Machine learning → Neural networks | |
| dc.subject | Information systems → Multimedia information systems | |
| dc.subject | Multimedia and multimodal retrieval | |
| dc.title | PARC: A Two-Stage Multi-Modal Framework for Point Cloud Completion | en_US |
Files
Original bundle
1 - 1 of 1