Research map / 3D vision / Face and shape
Pix2Shape: Towards Unsupervised Learning of 3D Scenes from Images Using a View-Based Representation
Sai Rajeswar, Fahim Mannan, Florian Golemo, Jérôme Parent-Lévesque, David Vázquez, Derek Nowrouzezahrai, Aaron C. Courville
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 reconstruction loss to adversarial loss (ablation) | Pix2Shape · ShapeNet scenes reconstruction, 0 degrees | Chamfer distance: 1.927 -> 0.1120 | Improved |
| add adversarial loss to reconstruction loss (ablation) | Pix2Shape · ShapeNet scenes reconstruction, 80 degrees | Chamfer distance: 0.1925 -> 0.1433 | Improved |
| add adversarial loss to reconstruction loss (ablation) | Pix2Shape · ShapeNet scenes reconstruction, 0 degrees | Chamfer distance: 0.1066 -> 0.1120 | Hurt |
| add mutual information loss (ablation) | Pix2Shape · 3D-IQTT (1000 labels) | accuracy: 0.4921 -> 0.7020 | Improved |
| add mutual information loss (ablation) | Pix2Shape · 3D-IQTT (0 labels) | accuracy: 0.4224 -> 0.5526 | Improved |
| add adversarial loss to reconstruction plus mutual info (ablation) | Pix2Shape · 3D-IQTT (1000 labels) | accuracy: 0.6272 -> 0.7020 | Improved |
| add reconstruction loss to adversarial loss (ablation) | Pix2Shape · 3D-IQTT (1000 labels) | accuracy: 0.4454 -> 0.4921 | Improved |
| use Pix2Shape instead of Perspective Transformer Nets | PTN · Shape scenes reconstruction | Chamfer distance: 0.145 -> 0.103 | Improved |
| use Pix2Shape instead of PTN | PTN · ShapeNet scenes reconstruction | MSE-depth: 0.103 -> 0.053 | Improved |
| use Pix2Shape instead of Rezende et al. | Rezende et al. · 3D-IQTT (0 labels) | accuracy: 0.5202 -> 0.5519 | Improved |
| use Pix2Shape instead of Rezende et al. | Rezende et al. · 3D-IQTT (1,000 labels) | accuracy: 0.7001 -> 0.7012 | No clear effect |
| add labeled samples (0 to 1,000) | Pix2Shape · 3D-IQTT | accuracy: 0.5519 -> 0.7012 | Improved |
Closest papers by meaning
- Pix2Vox: Context-Aware 3D Reconstruction From Single and Multi-View Images (ICCV 2019 · similarity 0.90)
- Pix2Vox++: Multi-scale Context-aware 3D Object Reconstruction from Single and Multiple Images (IJCV 2020 · similarity 0.90)
- Object-X: Learning to Reconstruct Multi-Modal 3D Object Representations (NeurIPS 2025 · similarity 0.89)
- Pix3D: Dataset and Methods for Single-Image 3D Shape Modeling (CVPR 2018 · similarity 0.89)
- CoReNet: Coherent 3D Scene Reconstruction from a Single RGB Image (ECCV 2020 · similarity 0.89)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · 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.
