Research map / 3D vision / Face and shape
ShapeCodes: Self-Supervised Feature Learning by Lifting Views to Viewgrids
Dinesh Jayaraman, Ruohan Gao, Kristen Grauman
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 relative (observer-aligned) viewgrid targets instead of canonical alignment (ablation) | ShapeCodes · ModelNet seen viewgrid completion | MSE x1000: 4.689 -> 3.718 | Improved |
| add relative (observer-aligned) viewgrid targets instead of canonical alignment (ablation) | ShapeCodes · ModelNet unseen viewgrid completion | MSE x1000: 10.440 -> 7.005 | Improved |
| add relative (observer-aligned) viewgrid targets instead of canonical alignment (ablation) | ShapeCodes · ShapeNet unseen viewgrid completion | MSE x1000: 9.021 -> 6.811 | Improved |
| use relative-aligned viewgrid features instead of canonical-alignment features (ablation) | ShapeCodes · ModelNet unseen recognition | accuracy (%): 69.6 -> 71.2 | Improved |
| use relative-aligned viewgrid features instead of canonical-alignment features (ablation) | ShapeCodes · ShapeNet seen recognition | accuracy (%): 56.9 -> 57.7 | Improved |
| replace egomotion pair prediction with full viewgrid prediction | ShapeCodes · ModelNet seen recognition | accuracy (%): 56.1 -> 65.2 | Improved |
| replace egomotion pair prediction with full viewgrid prediction | ShapeCodes · ShapeNet unseen recognition | accuracy (%): 49.7 -> 54.8 | Improved |
| replace autoencoder (reconstruct own view) with viewgrid prediction | ShapeCodes · ModelNet seen recognition | accuracy (%): 52.5 -> 65.2 | Improved |
| replace context inpainting with viewgrid prediction | ShapeCodes · ShapeNet seen recognition | accuracy (%): 46.2 -> 57.7 | Improved |
| replace explicit voxel (3D-R2N2) encoder with implicit viewgrid encoder | ShapeCodes · ModelNet seen recognition | accuracy (%): 49.4 -> 65.2 | Improved |
| replace point-cloud (PointSetNet) encoder with implicit viewgrid encoder | ShapeCodes · ModelNet seen recognition | accuracy (%): 35.5 -> 65.2 | Improved |
| self-supervised ShapeCode instead of ImageNet-supervised VGG | ShapeCodes · ModelNet unseen recognition | accuracy (%): 64.9 -> 71.2 | Improved |
Closest papers by meaning
- Straight to Shapes: Real-Time Detection of Encoded Shapes (CVPR 2017 · similarity 0.90)
- Planes vs. Chairs: Category-Guided 3D Shape Learning without Any 3D Cues (ECCV 2022 · similarity 0.88)
- Multiview Aggregation for Learning Category-Specific Shape Reconstruction (NeurIPS 2019 · similarity 0.88)
- Self-Supervised Neural Articulated Shape and Appearance Models (CVPR 2022 · similarity 0.88)
- Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesis (NeurIPS 2026 · similarity 0.87)
- ShapeEmbed: a self-supervised learning framework for 2D contour quantification (NeurIPS 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.
