Research map / Label-efficient and robust learning / Test-time adaptation
TTA-COPE: Test-Time Adaptation for Category-Level Object Pose Estimation
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 |
|---|---|---|---|
| Use online pose-aware teacher-student ensemble and self-training | UDA-COPE on REAL275 semi-supervised evaluation · REAL275 semi-supervised | mAP IoU75: 34.4 -> 43.5 | Helped |
| Use online pose-aware teacher-student ensemble and self-training | UDA-COPE on REAL275 unsupervised evaluation · REAL275 unsupervised | mAP IoU50: 67.2 -> 69.1 | Helped |
| Replace entropy minimization with pose-aware TTA-COPE | category-level pose estimation · REAL275 semi-supervised | mAP IoU75: 33.7 -> 43.5 | Helped |
| Filter noisy pseudo-label points | mean-teacher pseudo-label TTA · REAL275 semi-supervised | mAP IoU75: 39.9 -> 41.1 | Helped |
| Use pose-aware output-level inlier confidence instead of output Softmax Max (ablation) | pose ensemble at inference · REAL275 semi-supervised | mAP 5° 2cm: 32.1 -> 32.8 | Helped |
| Use output-level inlier-confidence selection instead of input-level Softmax Max (ablation) | pose ensemble at inference · REAL275 semi-supervised | mAP IoU75: 40.9 -> 41.2 | Helped |
| Use input Softmax Average rather than input Argmax Match (ablation) | pose ensemble at inference · REAL275 semi-supervised | mAP IoU75: 27.5 -> 40.4 | Helped |
| Add self-training loss Ld (ablation) | output inlier-max inference ensemble · REAL275 semi-supervised | mAP IoU75: 41.2 -> 42.9 | Helped |
| Add self-training loss Ld (ablation) | output inlier-max pseudo-label ensemble · REAL275 semi-supervised | mAP IoU75: 41.3 -> 43.0 | Helped |
| Apply the pose ensemble for inference rather than only pseudo-label generation (ablation) | output-level inlier-max ensemble · REAL275 semi-supervised | mAP 5° 2cm: 31.6 -> 32.8 | Helped |
| Reduce update frequency to every 10 frames (ablation) | TTA-COPE · REAL275 semi-supervised | mAP IoU75: 43.5 -> 41.8 | Hurt |
| Reduce update frequency to every 20 frames (ablation) | TTA-COPE · REAL275 semi-supervised | mAP IoU75: 43.5 -> 41.1 | Hurt |
Closest papers by meaning
- Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation (ICLR 2024 · similarity 0.87)
- Fast Adaptation for Human Pose Estimation via Meta-Optimization (CVPR 2024 · similarity 0.87)
- CatFormer: Category-Level 6D Object Pose Estimation with Transformer (AAAI 2024 · similarity 0.85)
- Self-Bootstrapping for Versatile Test-Time Adaptation (ICML 2025 · similarity 0.85)
- What How and When Should Object Detectors Update in Continually Changing Test Domains? (CVPR 2024 · similarity 0.85)
- NOPE: Novel Object Pose Estimation from a Single Image (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.