Research map / 3D vision / Human pose estimation
Understanding the Limitations of CNN-Based Absolute Camera Pose Regression
Check 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 DenseVLAD image retrieval instead of pose regression | APR and RPR methods · Cambridge Landmarks and 7 Scenes | median position and orientation error | Mixed |
| Use 3D structure-based localization instead of pose regression | APR methods · Cambridge Landmarks and 7 Scenes | median position and orientation error: 0.06 vs 1.49 | Helped |
| Interpolate poses of retrieved images | DenseVLAD · TUM LSI | median position error: 1.08m -> 0.49m | Helped |
| Add densely sampled training images | PoseNet · Synthetic Shop Facade | median position error: 1.19 -> 0.68 | Helped |
| Add densely sampled training images | MapNet · Synthetic Shop Facade | median position error: 1.07 -> 0.33 | Helped |
| Increase training data by dense sampling | PoseNet and MapNet · Synthetic Shop Facade | median position error versus Active Search: 0.68 vs 0.01; 0.33 vs 0.01 | Hurt |
| Use MapNet+ or MapNet+PGO instead of DenseVLAD | RobotCar LOOP · RobotCar LOOP | position error | Helped |
| Scale to a larger scene | MapNet+ and MapNet+PGO · RobotCar FULL | position error | Hurt |
| Use MapNet instead of image retrieval with abundant training data | Synthetic Shop Facade · Synthetic Shop Facade | pose accuracy | Helped |
| Capture training poses on a line and test from diverging viewpoints | PoseNet and MapNet · limited-training-data scenes | generalization to test poses | Hurt |
| Use image retrieval instead of sequence-based VLocNet | DeepLoc · DeepLoc | median position error: 0.57m vs 0.68m | Helped |
Closest papers by meaning
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.90)
- Map-Relative Pose Regression for Visual Re-Localization (CVPR 2024 · similarity 0.89)
- Scene-agnostic Pose Regression for Visual Localization (CVPR 2025 · similarity 0.89)
- ConDo: Continual Domain Expansion for Absolute Pose Regression (AAAI 2025 · similarity 0.89)
- Neural Refinement for Absolute Pose Regression with Feature Synthesis (CVPR 2024 · similarity 0.88)
- HyperPose: Hypernetwork-Infused Camera Pose Localization and an Extended Cambridge Landmarks Dataset (CVPR 2025 · similarity 0.88)
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.