Research map / 3D vision / Face and shape
DECOR-GAN: 3D Shape Detailization by Conditional Refinement
Check 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 |
|---|---|---|---|
| add adversarial training rather than using reconstruction loss alone (ablation) | DECOR-GAN · ShapeNet chairs | Div-IOU: 0.325 -> 0.800 | Helped |
| add the generator mask (ablation) | DECOR-GAN · ShapeNet chairs | Cls-score: 0.672 -> 0.644 | Helped |
| use a dilated rather than strict generator mask (ablation) | DECOR-GAN · ShapeNet chairs | Loose-IOU: 0.587 -> 0.805 | Helped |
| add the discriminator mask (ablation) | DECOR-GAN · ShapeNet chairs | Cls-score: 0.843 -> 0.644 | Helped |
| use separate style-specific and global discriminator branches instead of the conditional discriminator with receptive field 1^3 (ablation) | DECOR-GAN · ShapeNet chairs | Div-IOU: 0.291 -> 0.800 | Helped |
| use separate style-specific and global discriminator branches instead of the conditional discriminator with receptive field 3^3 (ablation) | DECOR-GAN · ShapeNet chairs | Div-IOU: 0.197 -> 0.800 | Helped |
| increase the style-loss weight alpha (ablation) | DECOR-GAN · ShapeNet chairs, N=64 styles | Div-IOU: 0.163 -> 0.800 | Helped |
| increase the style-loss weight alpha (ablation) | DECOR-GAN · ShapeNet chairs, N=64 styles | Loose-IOU: 0.868 -> 0.805 | Hurt |
| remove the global GAN loss (ablation) | DECOR-GAN · ShapeNet chairs, N=64 styles | Div-IOU: 0.800 -> 0.950 | Helped |
| increase the number of detailed training styles from 16 to 64 (ablation) | DECOR-GAN · ShapeNet chairs | Loose-IOU: 0.840 -> 0.868 | Helped |
| apply Gaussian smoothing to the detailed training voxels (ablation) | DECOR-GAN · ShapeNet chairs | Div-IOU: 0.153 -> 0.800 | Helped |
| increase Gaussian filter sigma beyond 1.0 (ablation) | DECOR-GAN · ShapeNet chairs | Loose-IOU: 0.805 -> 0.614 | Hurt |
Closest papers by meaning
- Unsupervised 3D Shape Completion Through GAN Inversion (CVPR 2021 · similarity 0.90)
- Efficient Geometry-Aware 3D Generative Adversarial Networks (CVPR 2022 · similarity 0.89)
- Do 2D GANs Know 3D Shape? Unsupervised 3D Shape Reconstruction from 2D Image GANs (ICLR 2021 · similarity 0.89)
- What You See is What You GAN: Rendering Every Pixel for High-Fidelity Geometry in 3D GANs (CVPR 2024 · similarity 0.89)
- Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes With Deep Generative Networks (CVPR 2017 · similarity 0.88)
- Dual Mapping of 2D StyleGAN for 3D-Aware Image Generation and Manipulation (Student Abstract) (AAAI 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.