Research map / Point clouds / Shape and surface completion
Point Cloud Completion with Pretrained Text-to-Image Diffusion Models
Yoni Kasten, Ohad Rahamim, Gal Chechik
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SDS-Complete instead of ShapeFormer on out-of-domain objects | ShapeFormer · Redwood OOD (average) | Chamfer distance: 75.7 -> 36.6 | Improved |
| Use SDS-Complete instead of PoinTr on out-of-domain objects | PoinTr · Redwood OOD (average) | Chamfer distance: 74.3 -> 36.6 | Improved |
| Use SDS-Complete instead of best baseline on in-domain objects | Sinv · Redwood in-domain (average) | Chamfer distance: 27.7 -> 21.5 | Improved |
| Use SDS-Complete vs Sinv on executive chair (in-domain) | Sinv · Redwood executive chair | Chamfer distance: 20.6 -> 23.7 | Hurt |
| Use a generic prompt 'An object' instead of the full text (ablation) | SDS-Complete · Redwood 10 objects (average) | Chamfer distance: 30.5 -> 53.9 | Hurt |
| Use a generic prompt 'A thing' instead of the full text (ablation) | SDS-Complete · Redwood 10 objects (average) | Chamfer distance: 30.5 -> 51.2 | Hurt |
| Use class name prompt instead of detailed full text (ablation) | SDS-Complete · Redwood 10 objects (average) | Chamfer distance: 32.6 -> 30.5 | No clear effect |
| Use incorrect text description (table for chair or chair for table) (ablation) | SDS-Complete · Redwood (qualitative) | completion quality | Mixed |
| Remove the SDS loss (ablation) | SDS-Complete · Redwood 10 scans (qualitative) | completion quality | Improved |
| Replace SDF representation with density function (ablation) | SDS-Complete · Redwood 10 scans (qualitative) | adherence to partial input | Improved |
| Use camera sampling curriculum instead of random cameras (ablation) | SDS-Complete · Redwood 10 scans (qualitative) | completion consistency | Improved |
| Use SDS-Complete instead of ShapeFormer and PoinTr in user study | ShapeFormer · KITTI user study | best quality (%): 26.6 -> 73.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PoinTr — PoinTr: Diverse Point Cloud Completion With Geometry-Aware Transformers (ICCV 2021) · better on 11 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Position-Aware Guided Point Cloud Completion with CLIP Model (AAAI 2025 · similarity 0.89)
- PCDreamer: Point Cloud Completion Through Multi-view Diffusion Priors (CVPR 2025 · similarity 0.88)
- ComPC: Completing a 3D Point Cloud with 2D Diffusion Priors (ICLR 2025 · similarity 0.88)
- GenPC: Zero-shot Point Cloud Completion via 3D Generative Priors (CVPR 2025 · similarity 0.88)
- Scaling Diffusion Models to Real-World 3D LiDAR Scene Completion (CVPR 2024 · similarity 0.88)
- Point-based Instance Completion with Scene Constraints (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
