Research map / 3D vision / Face and shape
FPNN: Field Probing Neural Networks for 3D Data
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train the field-probing locations and weights instead of keeping them fixed (ablation) | FPNN, 1-FC · ModelNet40 | top-1 accuracy: 79.1 to 85.0 | Helped |
| Train the field-probing locations and weights instead of keeping them fixed (ablation) | FPNN, 4-FCs · ModelNet40 | top-1 accuracy: 86.6 to 87.5 | Helped |
| Add normal fields to distance-field input (ablation) | FPNN, 1-FC · ModelNet40 | top-1 accuracy: 85.0 to 86.0 | Helped |
| Add normal fields to distance-field input (ablation) | FPNN, 4-FCs · ModelNet40 | top-1 accuracy: 87.5 to 88.4 | Helped |
| Use four fully connected layers rather than one (ablation) | FPNN with field probing · ModelNet40 | top-1 accuracy: 85.0 to 87.5 | Helped |
| Use shorter filter spans of 0.2-0.4 rather than 0.2-0.8 (ablation) | FPNN, 1-FC · ModelNet40 | accuracy: 85.0 to 84.1 | Hurt |
| Use shorter filter spans of 0.1-0.2 rather than 0.2-0.8 (ablation) | FPNN, 1-FC · ModelNet40 | accuracy: 85.0 to 82.8 | Hurt |
| Use shorter filter spans of 0.2-0.4 rather than 0.2-0.8 (ablation) | FPNN, 4-FCs · ModelNet40 | accuracy: 87.5 to 86.8 | Hurt |
| Use shorter filter spans of 0.1-0.2 rather than 0.2-0.8 (ablation) | FPNN, 4-FCs · ModelNet40 | accuracy: 87.5 to 86.9 | Hurt |
| Reduce the Gaussian transform's sigma below the selected value (ablation) | FPNN Gaussian layer · ModelNet40 | accuracy: ≈1% | Hurt |
| Train with field probing rather than using only a fully connected network (ablation) | FPNN, 1-FC · ModelNet40 | accuracy: 79.1 to 85.0 | Helped |
| Transfer learned field-probing features and fine-tune the target FC layer rather than train FC Only (ablation) | FPNN transfer to MN40_1 · MN40_1 | accuracy: 90.7% < 92.7% | Helped |
Closest papers by meaning
- Volumetric and Multi-View CNNs for Object Classification on 3D Data (CVPR 2016 · similarity 0.88)
- Neural Processing of Tri-Plane Hybrid Neural Fields (ICLR 2024 · similarity 0.87)
- Structured Receptive Fields in CNNs (CVPR 2016 · similarity 0.87)
- Learning to Infer Implicit Surfaces without 3D Supervision (NeurIPS 2019 · similarity 0.87)
- ARO-Net: Learning Implicit Fields From Anchored Radial Observations (CVPR 2023 · similarity 0.86)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.85)
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.