Research map / 3D vision / Depth and stereo
ViewNet: Unsupervised Viewpoint Estimation From Conditional Generation
Octave Mariotti, Oisin Mac Aodha, Hakan Bilen
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 cycle consistency with rendered novel views | ViewNet · ShapeNet airplane | accuracy (%): 82 -> 86 | Improved |
| add cycle consistency with rendered novel views | ViewNet · ShapeNet car | accuracy (%): 89 -> 91 | Improved |
| add cycle consistency with rendered novel views | ViewNet · ShapeNet chair | accuracy (%): 89 -> 92 | Improved |
| add multiple viewpoint heads (3 instead of 1) (ablation) | ViewNet · ShapeNet car | accuracy (%): 51 -> 89 | Improved |
| add multiple viewpoint heads (3 instead of 1) (ablation) | ViewNet · ShapeNet airplane | accuracy (%): 72 -> 82 | Improved |
| reconstruct pixel values instead of binary segmentation target (ablation) | ViewNet · ShapeNet airplane | accuracy (%): 71 -> 82 | Improved |
| add Gaussian shape prior (ablation) | ViewNet · ShapeNet chair | accuracy (%): 73 -> 89 | Improved |
| add Gaussian shape prior (ablation) | ViewNet · ShapeNet car | accuracy (%): 89 -> 89 | No clear effect |
| add AdaIN conditioning instead of plain encoder-decoder (ablation) | ViewNet · ShapeNet airplane | accuracy (%): 82 -> 82 | No clear effect |
| use analytic perspective renderer instead of learnable HoloGAN-style decoder (ablation) | ViewNet · ShapeNet car | accuracy (%): 52 -> 89 | Improved |
| add cycle consistency | ViewNet · PASCAL3D+ car | median error (deg): 5.9 -> 5.6 | Improved |
| add cycle consistency | ViewNet · PASCAL3D+ (trained on Freiburg Cars) | accuracy (%): 61 -> 59 | Hurt |
Closest papers by meaning
- Self-Supervised Viewpoint Learning From Image Collections (CVPR 2020 · similarity 0.89)
- AUTO3D: Novel view synthesis through unsupervisely learned variational viewpoint and global 3D representation (ECCV 2020 · similarity 0.89)
- Novel Object Viewpoint Estimation Through Reconstruction Alignment (CVPR 2020 · similarity 0.88)
- Perspective Transformer Nets: Learning Single-View 3D Object Reconstruction without 3D Supervision (NeurIPS 2016 · similarity 0.87)
- Learning View Priors for Single-View 3D Reconstruction (CVPR 2019 · similarity 0.87)
- One View Is Enough: In-the-Wild Monocular Pretraining for Novel View Generation (NeurIPS 2026 · similarity 0.86)
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.
