Research map / Point clouds / Shape and surface completion
Neural Compression of 3D Meshes using Sparse Implicit Representation
Jianqiang Wang, Siyu Ren, Junhui Hou
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 |
|---|---|---|---|
| represent meshes as sparse SDF tensors and compress with SNC rather than point-cloud or mesh codecs | SIR-SNC · Mixed (600 meshes) | BD-BR (CD): -55.8% vs G-PCC, -57.4% vs Draco, -39.2% vs NeCGS | Improved |
| increase residual blocks from 1 to 5 (ablation) | SIR-SNC · Mixed | BD-BR: anchor -> -25% | Improved |
| increase residual blocks from 5 to 7 (ablation) | SIR-SNC · Mixed | BD-BR: -25% -> -28%, params 430K -> 970K | No clear effect |
| increase feature channels from 16 to 32 (ablation) | SIR-SNC · Mixed | BD-BR | No clear effect |
| add occupancy (BCE) loss with voxel pruning (ablation) | SIR-SNC · Mixed | reconstruction quality: approximately 47% | Improved |
| raise occupancy loss weight alpha from 0.01 to 0.1 (ablation) | SIR-SNC · Mixed | reconstruction quality: 2% variation | No clear effect |
| partition data into independent 8x8x8 blocks instead of full-volume sparse convolution (ablation) | SIR-SNC (5 ResNet) · Mixed | RD performance: performance drop of more than 80% | Hurt |
| raise sparsity threshold tau above about 0.75 voxel diagonal (ablation) | SIR · Mixed | CD | No clear effect |
| overfit decoder to dynamic sequence and send parameters in bitstream | SIR-SNC · MPEG dynamic sequences | bitrate: >28% bitrate reduction | Improved |
Closest papers by meaning
- Neural Compression for 3D Geometry Sets (ICCV 2025 · similarity 0.91)
- Neural Unsigned Distance Fields for Implicit Function Learning (NeurIPS 2020 · similarity 0.87)
- MeshFormer : High-Quality Mesh Generation with 3D-Guided Reconstruction Model (NeurIPS 2024 · similarity 0.87)
- High Resolution UDF Meshing via Iterative Networks (NeurIPS 2025 · similarity 0.86)
- Sparc3D: Sparse Representation and Construction for High-Resolution 3D Shapes Modeling (NeurIPS 2025 · similarity 0.86)
- Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention (NeurIPS 2025 · similarity 0.86)
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.
