Research map / 3D vision / Face and shape
From Images to 3D Shape Attributes
David F. Fouhey, Abhinav Gupta, Andrew Zisserman
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 |
|---|---|---|---|
| switch base representation from VGG-M to VGG-16 | 3D shape attribute network · Sculptures test | mean AUROC: 72.3 -> 74.4 | Improved |
| evaluate only on most confident predictions (top and bottom 25%) | attribute network · Sculptures test | mean AUROC: 72.3 -> 77.9 | Improved |
| evaluate only on top and bottom 10% of predictions | attribute network · Sculptures test | mean AUROC: 72.3 -> 82.6 | Improved |
| add the shape embedding loss to attribute-only training (ablation) | attribute network · Mental rotation (Easy) | AUROC: +3.4% over best baseline (easy), +6.9% (hard) | Improved |
| train attribute loss on top of pretrained weights (attribute-only vs pretrained) (ablation) | attribute-only network · Mental rotation | AUROC: +0.9% easy, +2.5% hard | Improved |
| use 12D attributes instead of IFV representation | attribute representation · Mental rotation | AUROC: 3-4% better | Improved |
| vary shading cue only vs contour cue only on a sphere/cube (ablation) | attribute network · synthetic planarity probe | correlation r_s: 0.96 and 0.97 (varying shading) vs 0.50 and 0.36 (varying contour) | Mixed |
| replace FC layers with global average pooling (CAM network) (ablation) | attribute network · Sculptures test | mean AUROC: 0.64% lower | No clear effect |
| clean metadata (reassign works) | mental rotation evaluation · Mental rotation | AUROC: about 0.1% increase | No clear effect |
| use direct holistic attribute prediction instead of depth-then-attribute baselines | attribute network · Sculptures test | mean AUROC: 62.5 -> 72.3 | Improved |
Closest papers by meaning
- 3D Shape Attributes (CVPR 2016 · similarity 0.98)
- Learning to Predict 3D Surfaces of Sculptures from Single and Multiple Views (IJCV 2018 · similarity 0.91)
- SurfNet: Generating 3D Shape Surfaces Using Deep Residual Networks (CVPR 2017 · similarity 0.91)
- Image-Based 3D Object Reconstruction: State-of-the-Art and Trends in the Deep Learning Era (TPAMI 2019 · similarity 0.90)
- Survey and Evaluation of Neural 3D Shape Classification Approaches (TPAMI 2021 · similarity 0.90)
- Do 2D GANs Know 3D Shape? Unsupervised 3D Shape Reconstruction from 2D Image GANs (ICLR 2021 · similarity 0.90)
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.
