Research map / 3D vision / Depth and stereo
Visual Camera Re-Localization from RGB and RGB-D Images Using DSAC
Eric Brachmann, Carsten Rother
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 geometric training data augmentation (rotation, scaling) (ablation) | DSAC* · 7Scenes | accuracy: +9.1%, +7.7% and 4.1% | Improved |
| add data augmentation, RGB-only training (ablation) | DSAC* · 7Scenes Stairs | accuracy: +51.5% | Improved |
| add data augmentation, RGB-only training (ablation) | DSAC* · 12Scenes | accuracy: +8.6% | Improved |
| add data augmentation (ablation) | DSAC* · Cambridge Landmarks | median error | No clear effect |
| train VGGNet with DSAC* schedule instead of DSAC++ schedule (ablation) | DSAC++ architecture · 7Scenes (RGB + 3D model) | accuracy: 74.4% -> 82.0% | Improved |
| replace VGGNet with ResNet-style architecture under DSAC* training (ablation) | DSAC* · 7Scenes (RGB + 3D model) | accuracy: 82.0% -> 85.2% | Improved |
| use unified DSAC* initialization and network for RGB-only training | DSAC++ · Indoor RGB only | accuracy: 53.1% -> 80.7% | Improved |
| train without 3D model on Great Court instead of with SfM 3D model (ablation) | DSAC* · Cambridge Great Court | median translation error (cm): 49/0.3 -> 34/0.2 | Improved |
| increase receptive field from 81px to 149px (ablation) | DSAC* · 7Scenes | accuracy | Hurt |
| decrease receptive field from 81px to 49px (ablation) | DSAC* · 7Scenes | accuracy | Hurt |
| use global receptive field (encoder-decoder) instead of local patches (ablation) | DSAC* · 7Scenes | accuracy | Hurt |
| add end-to-end training after initialization (ablation) | DSAC* · Cambridge Landmarks | median pose error: decreases by ca. 30% | Improved |
Closest papers by meaning
- Learning Camera Localization via Dense Scene Matching (CVPR 2021 · similarity 0.88)
- Hierarchical Scene Coordinate Classification and Regression for Visual Localization (CVPR 2020 · similarity 0.88)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.87)
- On the Limits of Pseudo Ground Truth in Visual Camera Re-Localisation (ICCV 2021 · similarity 0.87)
- Learning Less Is More - 6D Camera Localization via 3D Surface Regression (CVPR 2018 · similarity 0.87)
- Accelerated Coordinate Encoding: Learning to Relocalize in Minutes Using RGB and Poses (CVPR 2023 · similarity 0.86)
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.
