Research map / 3D vision / Face and shape
Generating 3D Faces using Convolutional Mesh Autoencoders
Anurag Ranjan, Timo Bolkart, Soubhik Sanyal, Michael J. Black
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 |
|---|---|---|---|
| replace the PCA model with the convolutional mesh autoencoder (8-d latent) | expression/face reconstruction · interpolation experiment | mean Euclidean error (mm): 1.639 -> 0.845 | Improved |
| replace the PCA model with the convolutional mesh autoencoder | model size · interpolation experiment | number of parameters: 120,552 -> 33,856 | Improved |
| use CoMA instead of PCA on unseen expressions | face reconstruction · extrapolation experiment (bareteeth) | mean Euclidean error (mm): 1.957 -> 1.376 | Improved |
| use CoMA instead of FLAME on unseen expressions | face reconstruction · extrapolation experiment (bareteeth) | mean Euclidean error (mm): 2.002 -> 1.376 | Improved |
| use CoMA instead of PCA on unseen expressions | vertices within 1 mm · extrapolation experiment | vertices within 1 mm error (%): 45% -> 63.8% | Improved |
| use CoMA instead of PCA in interpolation | vertices within 1 mm · interpolation experiment | vertices within 1 mm error (%): 47.3% -> 72.6% | Improved |
| replace FLAME's PCA expression space with CoMA (DeepFLAME), 2 latent dims | FLAME · expression reconstruction | mean Euclidean error (mm): 0.668 -> 0.610 | Improved |
| replace FLAME's PCA expression space with CoMA (DeepFLAME), 8 latent dims | FLAME · expression reconstruction | mean Euclidean error (mm): 0.477 -> 0.432 | Improved |
| replace FLAME's PCA expression space with CoMA (DeepFLAME), 16 latent dims | FLAME · expression reconstruction | mean Euclidean error (mm): 0.351 -> 0.372 | Hurt |
| replace FLAME's PCA expression space with CoMA (DeepFLAME), 16 latent dims | FLAME · expression reconstruction | median Euclidean error (mm): 0.139 -> 0.125 | Improved |
| add the variational KL loss (w_kld=0.001) to CoMA (ablation) | CoMA · latent sampling with Gaussian noise | sample diversity (qualitative) | Improved |
Closest papers by meaning
- FACE: A Face-based Autoregressive Representation for High-Fidelity and Efficient Mesh Generation (CVPR 2026 · similarity 0.91)
- Learning Complete 3D Morphable Face Models From Images and Videos (CVPR 2021 · similarity 0.91)
- High-Fidelity Monocular Face Reconstruction Based on an Unsupervised Model-Based Face Autoencoder (TPAMI 2018 · similarity 0.90)
- A Decoupled 3D Facial Shape Model by Adversarial Training (ICCV 2019 · similarity 0.90)
- 3DFaceGAN: Adversarial Nets for 3D Face Representation, Generation, and Translation (IJCV 2020 · similarity 0.90)
- Building 3D Generative Models from Minimal Data (IJCV 2023 · similarity 0.90)
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.
