Research map / 3D vision / Face and shape
SHIC: Shape-Image Correspondences with no Keypoint Supervision
Aleksandar Shtedritski, Christian Rupprecht, Andrea Vedaldi
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 |
|---|---|---|---|
| use max pooling instead of mean pooling across template views (ablation) | SHIC · DensePose-LVIS (average) | geodesic error: 26.6 -> 24.9 | Improved |
| add synthetic (depth-conditioned diffusion) template images as supervision (ablation) | SHIC · DensePose-LVIS (average) | geodesic error: 25.6 -> 24.9 | Improved |
| add natural images to training instead of synthetic data only (ablation) | SHIC · DensePose-LVIS (average) | geodesic error: 33.8 -> 24.9 | Improved |
| add SD-DINO pseudo ground truth for natural images (ablation) | SHIC · DensePose-LVIS (average) | geodesic error: 65.8 -> 24.9 | Improved |
| add L_pseudo classification loss (ablation) | SHIC · DensePose-LVIS (average) | geodesic error: 27.3 -> 24.9 | Improved |
| add distance-aware loss L_dist (ablation) | SHIC · DensePose-LVIS (average) | geodesic error: 25.8 -> 24.9 | Improved |
| add both symmetry losses L_eq and L_cyc (ablation) | SHIC · DensePose-LVIS (average) | geodesic error: 26.1 -> 24.9 | Improved |
| render surface normals instead of shaded template (ablation) | SHIC pseudo-GT · DensePose-LVIS (average) | geodesic error: 25.2 -> 24.9 | Improved |
| add Stable Diffusion features to DINO features for matching (ablation) | SHIC pseudo-GT · DensePose-LVIS (average) | geodesic error: 25.4 -> 24.9 | Improved |
| increase number of training images per class from 50 to 2k+ (ablation) | SHIC · DensePose-LVIS | geodesic error: 34.7 -> 22.3 | Improved |
| train with unsupervised SHIC rather than supervised CSE on giraffe | SHIC · DensePose-LVIS Giraffe | geodesic error: 18.0 -> 18.1 | No clear effect |
Closest papers by meaning
- Canonical Surface Mapping via Geometric Cycle Consistency (ICCV 2019 · similarity 0.88)
- Articulation-Aware Canonical Surface Mapping (CVPR 2020 · similarity 0.86)
- Cyclic Functional Mapping: Self-supervised Correspondence between Non-isometric Deformable Shapes (ECCV 2020 · similarity 0.86)
- Continuous Surface Embeddings (NeurIPS 2020 · similarity 0.85)
- Unsupervised 3D Structure Inference from Category-Specific Image Collections (CVPR 2024 · similarity 0.85)
- Learning SO(3)-Invariant Semantic Correspondence via Local Shape Transform (CVPR 2024 · similarity 0.85)
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.
