Research map / Point clouds / Shape and surface completion
SA-ConvONet: Sign-Agnostic Optimization of Convolutional Occupancy Networks
Jiapeng Tang, Jiabao Lei, Dan Xu, Feiying Ma, Kui Jia, Lei Zhang
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 |
|---|---|---|---|
| optimize the whole network instead of only the encoder (ablation) | SA-ConvONet · ShapeNet-chair | Chamfer Distance: 0.516 -> 0.522 | No clear effect |
| optimize the whole network instead of only the encoder (ablation) | SA-ConvONet · Synthetic Room | Chamfer Distance: 0.516 -> 0.495 | Improved |
| optimize the whole network instead of only the encoder (ablation) | SA-ConvONet · ScanNet | Chamfer Distance: 0.741 -> 0.728 | Improved |
| add convolutional occupancy pre-training instead of SAL geometric (sphere) initialization (ablation) | SA-ConvONet · ShapeNet-chair (qualitative) | reconstruction quality | Improved |
| reduce input from 30,000 to 5,000 points (ablation) | SA-ConvONet · ShapeNet-chair | Chamfer Distance: 0.522 -> 0.529 | No clear effect |
| reduce input from 30,000 to 5,000 points (ablation) | SA-ConvONet · Synthetic Room | Chamfer Distance: 0.495 -> 0.511 | No clear effect |
| increase input to 50,000 points (ablation) | SA-ConvONet · Synthetic Room | Chamfer Distance: 0.495 -> 0.476 | No clear effect |
| sign-agnostic test-time optimization of the whole network vs CONet without optimization | SA-ConvONet · ShapeNet-chair | Chamfer Distance: 0.821 -> 0.522 | Improved |
| sign-agnostic test-time optimization vs CONet | SA-ConvONet · Synthetic Room | Chamfer Distance: 2.020 -> 0.495 | Improved |
| sign-agnostic test-time optimization vs CONet | SA-ConvONet · ScanNet | Chamfer Distance: 1.559 -> 0.728 | Improved |
| sign-agnostic optimization vs SPSR (needs normals) | SA-ConvONet · ScanNet | F-score (tau): 82.33 -> 82.08 | Hurt |
Closest papers by meaning
- Sign-Agnostic Implicit Learning of Surface Self-Similarities for Shape Modeling and Reconstruction From Raw Point Clouds (CVPR 2021 · similarity 0.92)
- A Survey and Benchmark of Automatic Surface Reconstruction From Point Clouds (TPAMI 2024 · similarity 0.89)
- Surface Reconstruction From Point Clouds: A Survey and a Benchmark (TPAMI 2024 · similarity 0.89)
- Unsupervised Occupancy Learning from Sparse Point Cloud (CVPR 2024 · similarity 0.89)
- OffsetOPT: Explicit Surface Reconstruction without Normals (CVPR 2025 · similarity 0.87)
- Occlusion-aware Non-Rigid Point Cloud Registration via Unsupervised Neural Deformation Correntropy (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.
