Research map / 3D vision / Face and shape
Inferring 3D Shapes from Image Collections Using Adversarial Networks
Matheus Gadelha, Aartika Rai, Subhransu Maji, Rui Wang
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 projection module and train on images instead of 2D-GAN generator | PrGAN · ModelNet chairs image generation | MMD (lower better): 90.13 -> 88.31 | Improved |
| train on 2D images only instead of 3D voxel data | PrGAN vs 3D-GAN · ModelNet chairs voxel generation | MMD (lower better): 347.55 -> 442.98 | Hurt |
| add viewpoint annotation via conditional discriminator | PrGAN silhouette · ShapeNet airplanes | average Chamfer IoU similarity: 0.421 -> 0.435 | Improved |
| use depth maps instead of silhouettes | PrGAN · ShapeNet airplanes | average Chamfer IoU similarity: 0.421 -> 0.472 | Improved |
| use part segmentation instead of silhouettes | PrGAN · ShapeNet airplanes | average Chamfer IoU similarity: 0.421 -> 0.502 | Improved |
| use part segmentation instead of depth | PrGAN · ShapeNet airplanes | G(z)->D similarity: 0.448 -> 0.507 | Improved |
| use part segmentation instead of depth | PrGAN · ShapeNet airplanes | D->G(z) similarity (variety): 0.497 -> 0.496 | No clear effect |
| add viewpoint annotation | PrGAN silhouette · ShapeNet airplanes | D->G(z) similarity: 0.442 -> 0.439 | No clear effect |
| increase number of views per object from 1 to 8 in training data | PrGAN · ModelNet chairs qualitative | shape quality: 1, 2, 4, 8 views | Improved |
Closest papers by meaning
- Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling (NeurIPS 2016 · similarity 0.93)
- Do 2D GANs Know 3D Shape? Unsupervised 3D Shape Reconstruction from 2D Image GANs (ICLR 2021 · similarity 0.92)
- Synthesizing 3D Shapes From Silhouette Image Collections Using Multi-Projection Generative Adversarial Networks (CVPR 2019 · similarity 0.91)
- Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes With Deep Generative Networks (CVPR 2017 · similarity 0.91)
- 3D generation on ImageNet (ICLR 2023 · similarity 0.90)
- 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.
