Research map / 3D vision / Human pose estimation
6D Camera Relocalization in Ambiguous Scenes via Continuous Multimodal Inference
Mai Bui, Tolga Birdal, Haowen Deng, Shadi Albarqouni, Leonidas Guibas, Slobodan Ilic, Nassir Navab
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 RWTA multi-hypothesis training instead of EWTA (k=50) (ablation) | Bingham mixture, K=50 · Ambiguous real scenes (average of 5) | ratio of correct poses @ 15deg/0.2m: 0.34 -> 0.56 | Improved |
| use RWTA instead of EWTA (k=25) (ablation) | Bingham mixture, K=50 · Ambiguous real scenes (average of 5) | ratio of correct poses @ 15deg/0.2m: 0.40 -> 0.56 | Improved |
| use multi-hypothesis Ours-RWTA instead of MC-Dropout (ablation) | pose regression · Ambiguous real scenes (average of 5) | ratio of correct poses @ 15deg/0.2m: 0.39 -> 0.56 | Improved |
| use Ours-RWTA instead of Bingham-MDN (mode collapse) (ablation) | Bingham mixture · Ambiguous real scenes (average of 5) | ratio of correct poses @ 15deg/0.2m: 0.28 -> 0.56 | Improved |
| use Ours-RWTA instead of MC-Dropout for finding dining table modes (ablation) | pose regression · Synthetic dining table | mode detection accuracy: 50% -> 96% | Improved |
| use Ours-RWTA instead of MC-Dropout for finding round table modes (ablation) | pose regression · Synthetic round table | mode detection accuracy: 24.8% -> 99.1% | Improved |
| use Ours-RWTA with each backbone instead of baselines (ablation) | Ours-RWTA · Backbones ResNet-18/34/50/Inception-v3 | ratio of correct poses @ 15deg/0.2m: 0.39 -> 0.56 | Improved |
| replace ResNet-34 with Inception-v3 backbone (ablation) | Ours-RWTA · Ambiguous real scenes (average of 5) | ratio of correct poses @ 15deg/0.2m: 0.56 -> 0.49 | Hurt |
| increase number of hypotheses K to very high values (ablation) | Ours-RWTA · Ambiguous real scenes (average of 5) | ratio of correct poses | Hurt |
| model rotations with continuous 6D representation instead of Bingham quaternion (ablation) | Ours-RWTA · Ambiguous real scenes (average of 5) | ratio of correct poses @ 15deg/0.2m: 0.56 -> 0.53 | No clear effect |
| construct Bingham matrix V with Birdal et al. method instead of Gram-Schmidt (ablation) | Ours-RWTA · Ambiguous real scenes (average of 5) | ratio of correct poses @ 20deg/0.3m: 0.50 -> 0.68 | Improved |
| construct V with skew-symmetric instead of Birdal et al. for Bingham-MDN (ablation) | Bingham-MDN · Ambiguous real scenes (average of 5) | ratio of correct poses @ 15deg/0.2m: 0.28 -> 0.38 | Improved |
Closest papers by meaning
- Scene Coordinate Reconstruction: Posing of Image Collections via Incremental Learning of a Relocalizer (ECCV 2024 · similarity 0.89)
- Accelerated Coordinate Encoding: Learning to Relocalize in Minutes Using RGB and Poses (CVPR 2023 · similarity 0.88)
- Deep Bingham Networks: Dealing with Uncertainty and Ambiguity in Pose Estimation (IJCV 2022 · similarity 0.88)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.87)
- Living Scenes: Multi-object Relocalization and Reconstruction in Changing 3D Environments (CVPR 2024 · similarity 0.86)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · 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.
