Research map / 3D vision / Face and shape
Learning Generative Models of Textured 3D Meshes From Real-World Images
Dario Pavllo, Jonas Kohler, Thomas Hofmann, Aurelien Lucchi
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 semantic template inference after silhouette optimization (ablation) | pose estimation, single template · Bird (CUB) pose | GD(1): 0.47 -> 0.29 | Improved |
| add semantic template inference after silhouette optimization (ablation) | pose estimation, single template · Car (P3D) pose | GD(1): 0.12 -> 0.11 | No clear effect |
| add semantic template inference after silhouette optimization (ablation) | pose estimation, multiple templates · Car (P3D) pose | GD(1): 0.10 -> 0.06 | Improved |
| use multiple mesh templates instead of one (ablation) | pose estimation with semantics · Bird (CUB) pose | GD(1): 0.29 -> 0.32 | Hurt |
| use multiple mesh templates instead of one (ablation) | pose estimation with semantics · Car (P3D) pose | GD(1): 0.11 -> 0.06 | Improved |
| repeat ambiguity detection and semantic inference a second time (ablation) | pose estimation, single template · Bird (CUB) pose | GD(1): 0.29 -> 0.29 | No clear effect |
| use full pipeline with semantics instead of silhouette-only poses (ablation) | generator, single template · CUB bird FID | FID: 73.67 -> 55.75 | Improved |
| use full pipeline with semantics instead of silhouette-only poses (ablation) | generator, single template · P3D car FID | FID: 38.16 -> 36.52 | Improved |
| use full pipeline with semantics instead of silhouette-only poses (ablation) | generator, single template · P3D airplane FID | FID: 100.5 -> 81.28 | Improved |
| use multiple templates instead of one (ablation) | generator with semantics · CUB bird FID | FID: 55.75 -> 56.54 | No clear effect |
| train one conditional generator for all classes (setting B) instead of per-class models (ablation) | generator, single template · ImageNet airplane FID | FID: 77.84 -> 46.46 | Improved |
| train one conditional generator for all classes (setting B) instead of per-class models (ablation) | generator, single template · ImageNet car FID | FID: 30.73 -> 39.68 | Hurt |
Closest papers by meaning
- Convolutional Generation of Textured 3D Meshes (NeurIPS 2020 · similarity 0.95)
- Leveraging 2D Data to Learn Textured 3D Mesh Generation (CVPR 2020 · similarity 0.93)
- GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images (NeurIPS 2022 · similarity 0.93)
- Fine Detailed Texture Learning for 3D Meshes With Generative Models (TPAMI 2023 · similarity 0.93)
- 3D generation on ImageNet (ICLR 2023 · similarity 0.91)
- What You See is What You GAN: Rendering Every Pixel for High-Fidelity Geometry in 3D GANs (CVPR 2024 · similarity 0.89)
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.
