Research map / 3D vision / Face and shape
Higher-Order Function Networks for Learning Composable 3D Object Representations
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 |
|---|---|---|---|
| encode each object by predicting all parameters of a small mapping network | HOF · ShapeNet 13-class subset | Average CD (×100): 1.549 (0.0117) -> 1.247 (0.0048) | Helped |
| use a per-sample function-parameter mapping instead of a fixed decoder | HOF-3 vs FoldingNet · ShapeNet 13-class subset | Average CD (×100): 1.549 (0.0117) -> 1.247 (0.0048) | Helped |
| use HOF rather than competing methods | ShapeNet 55-class reconstruction · ShapeNet 55 classes | F1: 0.290 -> 0.291 | Helped |
| replace the baseline encoder with ResNet18 (ablation) | HOF-3 · ShapeNet architectural variation | CD(P,T): 1.480 -> 1.529 | Hurt |
| replace ReLU with tanh in the decoder (ablation) | HOF-3 · ShapeNet architectural variation | CD(P,T): 1.480 -> 1.652 | Hurt |
| increase the decoder to 6 hidden layers (ablation) | HOF-3 · ShapeNet architectural variation | CD(P,T): 1.480 -> 1.499 | Hurt |
| sample from the sphere surface instead of the 3D sphere interior (ablation) | HOF-3 · ShapeNet canonical sampling | CD(T,P): 1.014 -> 1.263 | Hurt |
| sample from a 4D sphere instead of the 3D sphere interior (ablation) | HOF-3 · ShapeNet canonical sampling | CD(T,P): 1.014 -> 0.912 | Helped |
| sample from a 3D cube instead of the 3D sphere interior (ablation) | HOF-3 · ShapeNet canonical sampling | CD(P,T): 1.480 -> 1.481 | No clear effect |
| use two self-compositions (k=2) (ablation) | ResNet18 HOF-1 · ShapeNet | CD(P,T): 1.635 -> 1.574 | Helped |
| use three self-compositions (k=3) (ablation) | ResNet18 HOF-1 · ShapeNet | CD(P,T): 1.635 -> 1.601 | Helped |
| use four self-compositions (k=4) (ablation) | ResNet18 HOF-1 · ShapeNet | CD(P,T): 1.635 -> 1.655 | Hurt |
Closest papers by meaning
- Neural Parts: Learning Expressive 3D Shape Abstractions With Invertible Neural Networks (CVPR 2021 · similarity 0.90)
- Learning Unsupervised Hierarchical Part Decomposition of 3D Objects From a Single RGB Image (CVPR 2020 · similarity 0.89)
- GEOMetrics: Exploiting Geometric Structure for Graph-Encoded Objects (ICML 2019 · similarity 0.89)
- Perspective Transformer Nets: Learning Single-View 3D Object Reconstruction without 3D Supervision (NeurIPS 2016 · similarity 0.88)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.88)
- Object-X: Learning to Reconstruct Multi-Modal 3D Object Representations (NeurIPS 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.