Research map / 3D vision / Face and shape
Continuous Surface 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 |
|---|---|---|---|
| Replace IUV chart coordinates with continuous surface embeddings | DP-RCNN-DeepLab* (R101) on DensePose-COCO · DensePose-COCO minival | AP: 67.7 -> 68.0 | Helped |
| Use geodesically smoothed loss L_sigma instead of standard loss L (ablation) | CSE on DensePose-COCO · DensePose-COCO minival, 100% training data | AP: 65.4 -> 66.1 | Helped |
| Use geodesically smoothed loss L_sigma instead of loss L (ablation) | CSE on DensePose-Chimps · DensePose-Chimps | AP on chimp mesh: 8.5 -> 21.1 | Helped |
| Initialize the predictor with human-trained weights (ablation) | CSE class-agnostic animal training · DensePose-Chimps | AP on chimp mesh: 21.1 -> 36.7 | Helped |
| Align animal embeddings to the human embedding space (ablation) | CSE predictor initialized with human weights · DensePose-Chimps | AP on chimp mesh: 36.7 -> 37.2 | Helped |
| Align animal embeddings to the human embedding space (ablation) | CSE predictor initialized with human weights · DensePose-Chimps | AP on SMPL mesh: 34.5 -> 35.7 | Helped |
| Align animal embeddings to the human embedding space (ablation) | Joint class-agnostic CSE animal training with pretrained predictor · DensePose-LVIS mean | AP: 33.1 -> 35.0 | Helped |
| Use a shared class-agnostic output space rather than separate class-specific output planes (ablation) | Joint CSE animal training with L_sigma · DensePose-LVIS mean | AP: 25.0 -> 25.2 | Helped |
| Reduce LBO basis size from 256 to 32 (ablation) | CSE with D=16 and L_sigma · DensePose-COCO minival | AP: 66.1 -> 63.2 | Hurt |
| Increase LBO basis size from 256 to 512 (ablation) | CSE with D=16 and L_sigma · DensePose-COCO minival | AP: 66.1 -> 65.9 | No clear effect |
| Reduce embedding dimension from 16 to 2 (ablation) | CSE with M=256 and L_sigma · DensePose-COCO minival | AP: 66.1 -> 46.4 | Hurt |
| Increase embedding dimension from 16 to 32 (ablation) | CSE with M=256 and L_sigma · DensePose-COCO minival | AP: 66.1 -> 66.0 | No clear effect |
Closest papers by meaning
- Discovering Relationships Between Object Categories via Universal Canonical Maps (CVPR 2021 · similarity 0.91)
- Unsupervised Learning of Dense Shape Correspondence (CVPR 2019 · similarity 0.91)
- Canonical 3D Deformer Maps: Unifying parametric and non-parametric methods for dense weakly-supervised category reconstruction (NeurIPS 2020 · similarity 0.90)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.87)
- DenseMatcher: Learning 3D Semantic Correspondence for Category-Level Manipulation from a Single Demo (ICLR 2025 · similarity 0.87)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (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.