Research map / 3D vision / Face and shape
Do It Yourself: Learning Semantic Correspondence from Pseudo-Labels
Olaf Dünkel, Thomas Wimmer, Christian Theobalt, Christian Rupprecht, Adam Kortylewski
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 pseudo-label adapter training (naive labels) over zero-shot (ablation) | SD+DINOv2 · SPair-71k | PCK@0.1: 65.0 -> 67.2 | Improved |
| add exact cyclic consistency filtering (ablation) | pseudo-labels · SPair-71k | PCK@0.1: 67.2 -> 66.9 | Hurt |
| add relaxed cyclic consistency filtering (ablation) | pseudo-labels · SPair-71k | PCK@0.1: 67.2 -> 68.4 | Improved |
| add 3D-aware chaining (ablation) | pseudo-labels with relaxed cyclic consistency · SPair-71k | PCK@0.1: 68.4 -> 70.0 | Improved |
| add spherical prototype rejection (ablation) | pseudo-labels with relaxed cyclic consistency and chaining · SPair-71k | PCK@0.1: 70.0 -> 74.4 | Improved |
| add spherical rejection to naive pseudo-labels (ablation) | naive pseudo-labels · SPair-71k | PCK@0.1: 67.2 -> 72.9 | Improved |
| use ground-truth pose instead of Orient-Anything estimated pose (ablation) | DIY-SC · SPair-71k (per-image) | PCK@0.1: 69.6 -> 71.6 | Improved |
| use SAM masks instead of rembg masks (ablation) | DIY-SC with estimated pose · SPair-71k (per-image) | PCK@0.1: 68.0 -> 69.6 | Improved |
| generate pseudo-labels with SD+DINOv2 instead of DINOv2 only (ablation) | DINOv2 refiner · SPair-71k (per-image) | PCK@0.1: 67.17 -> 71.6 | Improved |
| add ImageNet-3D pretraining then fine-tune on SPair | DIY-SC · SPair-71k (per-image) | PCK@0.1: 71.6 -> 72.2 | Improved |
| reduce adapter output channels from 1536 to 128 (ablation) | DIY-SC · SPair-71k | PCK@0.1: 74.63 -> 74.08 | No clear effect |
| fine-tune adapter with ground-truth labels after pseudo-label pretraining | supervised TLR · SPair-71k (per-image, supervised) | PCK@0.1: 82.9 -> 83.5 | Improved |
Closest papers by meaning
- Improving Semantic Correspondence with Viewpoint-Guided Spherical Maps (CVPR 2024 · similarity 0.92)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (NeurIPS 2025 · similarity 0.88)
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 2025 · similarity 0.88)
- Bridging Viewpoint Gaps: Geometric Reasoning Boosts Semantic Correspondence (CVPR 2025 · similarity 0.88)
- Telling Left from Right: Identifying Geometry-Aware Semantic Correspondence (CVPR 2024 · similarity 0.88)
- Learning SO(3)-Invariant Semantic Correspondence via Local Shape Transform (CVPR 2024 · 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.
