Research map / 3D vision / 3D generation
Virtual Training for a Real Application: Accurate Object-Robot Relative Localization Without Calibration
Vianney Loing, Renaud Marlet, Mathieu Aubry
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 |
|---|---|---|---|
| Predict x, y and theta with separate 1D bins instead of joint 2D (x,y) bins (ablation) | coarse CNN · Coarse x, lab, 1 camera | success e_x <= 60mm: 6.1 -> 98.7 | Improved |
| Add theta and y heads to the 1D x network (joint training) (ablation) | coarse CNN · Coarse x, lab, 1 camera | success e_x <= 60mm: 97.0 -> 98.7 | Improved |
| Replace single-step localization with three-step virtual close-up (ablation) | pose CNN · Lab dataset, 1 camera | success e_x <= 5mm: 20.5 -> 61.1 | Improved |
| Replace single-step localization with three-step virtual close-up (ablation) | pose CNN · Lab dataset, 1 camera | joint success: 2.8 -> 36.9 | Improved |
| Aggregate 3 cameras and 2 clamp positions (ablation) | three-step · Lab dataset | joint success: 36.9 -> 79.0 | Improved |
| Aggregate 3 cameras for clamp P1 x estimate (ablation) | fine CNN · Lab dataset | x error (mm): 7.5 ± 16.0 -> 3.4 ± 7.6 | Improved |
| Aggregate both clamp orientations (ablation) | fine CNN, all cameras · Lab dataset | x error (mm): 3.4 ± 7.6 -> 2.3 ± 1.8 | Improved |
| Move from lab to field conditions (dirt, gravel) | three-step, 3 cameras 2 clamps · Fine pose | joint success: 79.9 -> 70.2 | Hurt |
| Move from lab to adversarial conditions (paper distractors) | three-step, 3 cameras 2 clamps · Fine pose | joint success: 79.9 -> 45.1 | Hurt |
| Move from synthetic to real lab images | fine CNN, 1 camera · Fine pose | success e_x <= 5mm: 85.3 -> 62.8 | Hurt |
| Move from synthetic to real lab images | fine CNN, 1 camera · Fine pose | success e_theta <= 2deg: 68.4 -> 83.2 | Improved |
| Use three-step with aggregation instead of ArUco marker on block | three-step · Lab dataset | success e_x <= 5mm: 88.4 -> 90.8 | No clear effect |
Closest papers by meaning
- Learning Less Is More - 6D Camera Localization via 3D Surface Regression (CVPR 2018 · similarity 0.87)
- Learning Where to Look: Self-supervised Viewpoint Selection for Active Localization using Geometrical Information (ECCV 2024 · similarity 0.86)
- Geometry-Aware Learning of Maps for Camera Localization (CVPR 2018 · similarity 0.86)
- Learning To Detect Scene Landmarks for Camera Localization (CVPR 2022 · similarity 0.85)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.84)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · 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.
