Research map / Label-efficient and robust learning / Knowledge distillation
S$$^{2}$$P$$^{3}$$: Self-Supervised Polarimetric Pose Prediction
Patrick Ruhkamp, Daoyi Gao, Nassir Navab, Benjamin Busam
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 self-supervision on real data (S2P3) to the synthetically pre-trained student | student network · PhoCal (mean of 4 objects) | ADD(-S) average recall: 52.9 -> 80.7 | Improved |
| use polarimetric self-supervision (RGB-P) instead of Self6D++ RGB-D self-supervision | S2P3 · PhoCal (mean of 4 objects) | ADD(-S) average recall: 34.0 -> 80.7 | Improved |
| add L_pose loss against teacher pseudo-labels (ablation) | S2P3 · PhoCal (mean of 4 objects) | ADD(-S) average recall: 2.5 -> 80.7 | Improved |
| add L_physics loss (ablation) | S2P3 · PhoCal (mean of 4 objects) | ADD(-S) average recall: 72.3 -> 80.7 | Improved |
| add L_normal loss (ablation) | S2P3 · PhoCal (mean of 4 objects) | ADD(-S) average recall: 72.7 -> 80.7 | Improved |
| add L_mask loss (ablation) | S2P3 · PhoCal (mean of 4 objects) | ADD(-S) average recall: 74.4 -> 80.7 | Improved |
| use lightweight student instead of large teacher-style student (ablation) | S2P3 with self-supervision · PhoCal (mean of 4 objects) | ADD(-S) average recall: 73.0 -> 80.7 | Improved |
| add self-supervision to the large student (ablation) | large student · PhoCal (mean of 4 objects) | ADD(-S) average recall: 72.7 -> 73.0 | Improved |
| use polarimetric L_physics instead of depth chamfer loss (ablation) | S2P3 · PhoCal (mean of 4 objects) | ADD(-S) average recall: 52.9 -> 80.7 | Improved |
| use polarimetric L_physics instead of pixel-wise depth loss (ablation) | S2P3 · PhoCal (mean of 4 objects) | ADD(-S) average recall: 58.0 -> 80.7 | Improved |
| use depth chamfer loss instead of polarimetric loss (ablation) | S2P3 · PhoCal (Cup) | ADD(-S) average recall: 93.8 -> 100.0 | Improved |
| add self-supervision to the student | student network · PhoCal (Fork) | ADD(-S) average recall: 64.4 -> 72.4 | No clear effect |
Closest papers by meaning
- Polarimetric Pose Prediction (ECCV 2022 · similarity 0.90)
- Deep Shape from Polarization (ECCV 2020 · similarity 0.87)
- PolarDepth: Monocular Transparent Object Depth from Polar-Physics Priors (ICML 2026 · similarity 0.86)
- Shape From Polarization for Complex Scenes in the Wild (CVPR 2022 · similarity 0.86)
- RGB-to-Polarization Estimation: A New Task and Benchmark Study (NeurIPS 2025 · similarity 0.85)
- Glossy Object Reconstruction with Cost-effective Polarized Acquisition (CVPR 2025 · similarity 0.84)
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.
