Research map / 3D vision / Face and shape
Cross-Domain 3D Equivariant Image Embeddings
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Map images into pretrained spherical CNN feature space and estimate pose by spherical cross-correlation rather than direct pose regression | Relative pose estimation · ObjectNet3D airplane and car | median error (degrees): 13.75 vs 36.52; 8.22 vs 16.16 | Helped |
| Map images into pretrained spherical CNN feature space and estimate pose by spherical cross-correlation rather than direct pose regression | Relative pose estimation · ShapeNet 3DOF IB and CB | median error (degrees): 6.64 vs 45.4; 3.84 vs 9.83; 5.55 vs 10.8; 5.21 vs 22.2; 7.27 vs 44.4; 4.59 vs 10.5; 12.3 vs 25.6; 9.66 vs 24.5 | Helped |
| Use Huber loss instead of L1 or L2 loss for embedding prediction (ablation) | Cross-domain spherical embedding · Relative pose estimation | median pose error (degrees): approximately 10 deg larger with L1 or L2 | Helped |
| Replace standard convolutional layers with residual layers (ablation) | Cross-domain embedding encoder-decoder · Embedding prediction | performance | Helped |
| Add skip connections to the cross-domain encoder-decoder (ablation) | Cross-domain embedding encoder-decoder · Embedding prediction | performance | Hurt |
| Train the synthesizer on predicted image embeddings f(y) instead of target spherical CNN embeddings s(x) (ablation) | Novel view synthesizer · Novel view synthesis | synthesis performance | Helped |
| Warm-start the 3DOF regression model from a network pretrained on the 2DOF training set (ablation) | Regression pose estimator · 3DOF pose estimation | performance | Helped |
| Regress relative pose from pairs of views instead of regressing canonical orientations (ablation) | Regression pose estimator · Relative pose estimation | performance | Hurt |
Closest papers by meaning
- Rotation Equivariant Graph Convolutional Network for Spherical Image Classification (CVPR 2020 · similarity 0.89)
- Spherical CNNs (ICLR 2018 · similarity 0.89)
- 3D Equivariant Pose Regression via Direct Wigner-D Harmonics Prediction (NeurIPS 2024 · similarity 0.87)
- Structuring Representation Geometry with Rotationally Equivariant Contrastive Learning (ICLR 2024 · similarity 0.87)
- Neural Isometries: Taming Transformations for Equivariant ML (NeurIPS 2024 · similarity 0.86)
- TetraSphere: A Neural Descriptor for O(3)-Invariant Point Cloud Analysis (CVPR 2024 · 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.