Research map / Point clouds / Shape and surface completion
Learning Implicit Functions for Dense 3D Shape Correspondence of Generic Objects
Feng Liu, Xiaoming Liu
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 |
|---|---|---|---|
| add uncertainty learning to the deep branched implicit function (ablation) | Ours · BHCP semantic correspondence (3 categories, threshold 0.05) | relative accuracy: 5.3% -> 9.2% | Improved |
| add uncertainty learning (ablation) | Ours · BHCP helicopter (unseen, plane model) | relative accuracy: 9.6% | Improved |
| add uncertainty learning (ablation) | Ours · chair non-existence detection | AUC (%): 96.58 -> 97.24 | Improved |
| proposed vs best SoTA baseline on semantic correspondence | Ours · BHCP 4 categories (aligned) | relative accuracy: 29.3% | Improved |
| deep branched implicit function with semantic embedding vs preliminary work | Ours · ShapeNet part reconstruction | CD-L1: 0.024 -> 0.017 | Improved |
| proposed vs BAE-Net | Ours · ShapeNet part reconstruction | CD-L1: 0.032 -> 0.017 | Improved |
| proposed vs BAE-Net in segmentation | Ours · ShapeNet part segmentation | IoU: 86.1 -> 87.5 | Improved |
| proposed vs preliminary work in segmentation | Ours · ShapeNet part segmentation | IoU: 88.0 -> 87.5 | No clear effect |
| use rotation unaligned test data instead of aligned | Ours · BHCP unaligned | accuracy | Mixed |
| add the cross-reconstruction CD and smoothness loss terms (ablation) | Ours · BHCP chair correspondence | accuracy | Improved |
| vary PEV dimensionality k (ablation) | Ours · Chair segmentation and correspondence | accuracy: k=12 best | Improved |
| add noise N(0,0.02^2) to test chairs | Ours · BHCP chair | accuracy | Hurt |
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.
- BAE-Net — BAE-NET: Branched Autoencoder for Shape Co-Segmentation (ICCV 2019) · better on 8 of 10 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
- Learning Implicit Functions for Topology-Varying Dense 3D Shape Correspondence (NeurIPS 2020 · similarity 0.98)
- Unsupervised Learning of Dense Shape Correspondence (CVPR 2019 · similarity 0.89)
- Learning 3D Dense Correspondence via Canonical Point Autoencoder (NeurIPS 2021 · similarity 0.89)
- Deep Functional Maps: Structured Prediction for Dense Shape Correspondence (ICCV 2017 · similarity 0.89)
- Learning Canonical Embeddings for Unsupervised Shape Correspondence With Locally Linear Transformations (TPAMI 2023 · similarity 0.89)
- Learning SO(3)-Invariant Semantic Correspondence via Local Shape Transform (CVPR 2024 · similarity 0.88)
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.
