Research map / 3D vision / Face and shape
Deforming Autoencoders: Unsupervised Disentangling of Shape and Appearance
Zhixin Shu, Mihir Sahasrabudhe, Riza Alp Guler, Dimitris Samaras, Nikos Paragios, Iasonas Kokkinos
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 integral deformation to affine deformation (ablation) | DAE · MAFL test (landmark error) | mean landmark error (% inter-ocular distance): 14.13 -> 8.50 | Improved |
| train the DAE on CelebA instead of MAFL | DAE A+I · MAFL test (landmark error) | mean landmark error (% inter-ocular distance): 8.50 -> 7.54 | Improved |
| train a landmark regressor on deformation fields | DAE A+I CelebA · MAFL test (landmark error) | mean landmark error (% inter-ocular distance): 7.54 -> 5.96 | Improved |
| add regularization on estimated deformations (ablation) | DAE latent 32 · MAFL test (landmark error) | mean landmark error (% inter-ocular distance): 10.24 -> 5.96 | Improved |
| use integral deformation instead of residual deformation grid (ablation) | DAE latent 32 · MAFL test (landmark error) | mean landmark error (% inter-ocular distance): 9.93 -> 5.96 | Improved |
| increase latent size from 16 to 96 (ablation) | DAE · MAFL test (landmark error) | mean landmark error (% inter-ocular distance): 5.71 -> 6.46 | Mixed |
| add dense connections and increase latent size (ablation) | DAE · MAFL test (landmark error) | mean landmark error (% inter-ocular distance): 6.46 -> 5.45 | Improved |
| increase Z_T dimension from 0 to moderate (4-32D) (ablation) | DAE · MAFL faces (qualitative) | texture alignment and appearance fidelity | Improved |
| increase Z_T to 128-D (ablation) | DAE · MAFL faces (qualitative) | texture alignment | Hurt |
| use affine+integral warping instead of direct convolutional warp field (ablation) | DAE · MAFL faces (qualitative) | reconstruction and texture quality | Improved |
Closest papers by meaning
- Unsupervised Part-Based Disentangling of Object Shape and Appearance (CVPR 2019 · similarity 0.89)
- Explicit Disentanglement of Appearance and Perspective in Generative Models (NeurIPS 2019 · similarity 0.88)
- Unsupervised Learning of Probably Symmetric Deformable 3D Objects From Images in the Wild (CVPR 2020 · similarity 0.88)
- Unsupervised Disentangling of Appearance and Geometry by Deformable Generator Network (CVPR 2019 · similarity 0.88)
- Improving Shape Deformation in Unsupervised Image-to-Image Translation (ECCV 2018 · similarity 0.88)
- Neural Implicit Morphing of Face Images (CVPR 2024 · 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.
