Research map / 3D vision / Face and shape
Learning a Predictable and Generative Vector Representation for Objects
Rohit Girdhar, David F. Fouhey, Mikel Rodriguez, Abhinav Gupta
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 joint fine-tuning of the full TL network (ablation) | TL-embedding network · ShapeNet CAD voxel prediction | Average AP: 62.7 -> 65.4 | Improved |
| add joint fine-tuning (ablation) | TL-embedding network · ShapeNet CAD voxel prediction (Bed) | AP: 33.8 -> 41.9 | Improved |
| add joint fine-tuning (ablation) | TL-embedding network · ShapeNet CAD voxel prediction (Sofa) | AP: 79.3 -> 79.3 | No clear effect |
| add joint fine-tuning (ablation) | autoencoder · ShapeNet CAD reconstruction | Overall AP: 97.6 -> 97.6 | No clear effect |
| predict 64-D embedding then decode instead of predicting voxels directly (Direct-conv4) (ablation) | AlexNet image to voxel · ShapeNet CAD voxel prediction | Average AP: 38.0 -> 65.4 | Improved |
| predict embedding instead of voxels directly (Direct-fc8) (ablation) | AlexNet image to voxel · ShapeNet CAD voxel prediction | Average AP: 24.8 -> 65.4 | Improved |
| predict embedding instead of voxels directly (ablation) | AlexNet image to voxel · IKEA voxel prediction | Overall AP: 31.1 -> 38.3 | Improved |
| remove translational alignment correction at evaluation | TL-embedding network · IKEA voxel prediction | AP: 33% vs 28% | Improved |
| use convolutional autoencoder instead of PCA with same dimensions (ablation) | voxel embedding · ShapeNet CAD reconstruction | Overall AP: 96.8 -> 97.6 | Improved |
| use autoencoder embedding instead of PCA for classification (ablation) | 64D features + linear SVM · ModelNet40 | accuracy: 68.4% -> 74.4% | Improved |
| use learned embedding retrieval instead of AlexNet fc7 nearest neighbor | CAD retrieval · IKEA CAD retrieval | Mean recall@10 overall: 19.4 -> 29.3 | Improved |
| use learned embedding retrieval instead of fc7 nearest neighbor | CAD retrieval · IKEA CAD retrieval (Table) | Mean recall@10: 17.4 -> 8.0 | Hurt |
Closest papers by meaning
- 3D-PRNN: Generating Shape Primitives With Recurrent Neural Networks (ICCV 2017 · similarity 0.87)
- LaGeM: A Large Geometry Model for 3D Representation Learning and Diffusion (ICLR 2025 · similarity 0.86)
- Unsupervised Learning of 3D Structure from Images (NeurIPS 2016 · similarity 0.86)
- Higher-Order Function Networks for Learning Composable 3D Object Representations (ICLR 2020 · similarity 0.86)
- DeepVoxels: Learning Persistent 3D Feature Embeddings (CVPR 2019 · similarity 0.86)
- Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling (NeurIPS 2016 · 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.
