Research map / 3D vision / Face and shape
DEMEA: Deep Mesh Autoencoders for Non-Rigidly Deforming Objects
Edgar Tretschk, Ayush Tewari, Michael Zollhöfer, Vladislav Golyanik, Christian Theobalt
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 |
|---|---|---|---|
| use regressed EDL rotations instead of graph loss (GL) (ablation) | DEMEA · DFaust (cm), latent dim 8 | average per-vertex error: 8.92 -> 6.69 | Improved |
| use regressed EDL rotations instead of graph loss (GL) (ablation) | DEMEA · CoMA (mm), latent dim 32 | average per-vertex error: 0.99 -> 0.81 | Improved |
| use regressed EDL rotations instead of local Procrustes (LP) (ablation) | DEMEA · DFaust (cm), latent dim 8 | average per-vertex error: 7.71 -> 6.69 | Improved |
| use regressed EDL rotations instead of local Procrustes (LP) (ablation) | DEMEA · Textureless cloth (mm), latent dim 32 | average per-vertex error: 7.96 -> 6.40 | Improved |
| use regressed EDL rotations instead of local Procrustes (LP) (ablation) | DEMEA · DFaust (cm), latent dim 32 | average per-vertex error: 2.22 -> 2.23 | No clear effect |
| add embedded deformation layer instead of learned upsampling (CA) (ablation) | DEMEA · SynHand5M (mm), latent dim 32 | average per-vertex error: 2.60 -> 2.51 | Improved |
| add embedded deformation layer instead of learned upsampling (CA) (ablation) | DEMEA · DFaust (cm), latent dim 32 | average per-vertex error: 2.07 -> 2.23 | Hurt |
| add embedded deformation layer instead of learned upsampling (CA) (ablation) | DEMEA · CoMA (mm), latent dim 32 | average per-vertex error: 0.72 -> 0.81 | Hurt |
| use graph-convolutional decoder with EDL instead of almost-linear FC network (FCA) (ablation) | DEMEA · SynHand5M (mm), latent dim 8 | average per-vertex error: 15.10 -> 8.12 | Improved |
| use graph-convolutional decoder with EDL instead of almost-linear FC network (FCA) (ablation) | DEMEA · Textureless cloth (mm), latent dim 8 | average per-vertex error: 15.63 -> 11.28 | Improved |
| use spiral convolutions instead of spectral convolutions (ablation) | DEMEA · SynHand5M (mm), latent dim 32 | average per-vertex error: 3.83 -> 2.51 | Improved |
| use spiral convolutions instead of spectral convolutions (ablation) | DEMEA · DFaust, latent dim 8 | average per-vertex error: 6.56 -> 6.69 | Mixed |
Closest papers by meaning
- Variational Autoencoders for Deforming 3D Mesh Models (CVPR 2018 · similarity 0.89)
- Mesh-Based Autoencoders for Localized Deformation Component Analysis (AAAI 2018 · similarity 0.89)
- Variational Autoencoders for Localized Mesh Deformation Component Analysis (TPAMI 2021 · similarity 0.89)
- Deformable Shape Completion With Graph Convolutional Autoencoders (CVPR 2018 · similarity 0.88)
- Fully Convolutional Mesh Autoencoder using Efficient Spatially Varying Kernels (NeurIPS 2020 · similarity 0.88)
- 4Deform: Neural Surface Deformation for Robust Shape Interpolation (CVPR 2025 · similarity 0.84)
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.
