Research map / 3D vision / Depth and stereo
Image-Based Localization Using LSTMs for Structured Feature Correlation
Florian Walch, Caner Hazirbas, Laura Leal-Taixe, Torsten Sattler, Sebastian Hilsenbeck, Daniel Cremers
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 LSTM-based structured dimensionality reduction after the FC layer (vs PoseNet) | PoseNet (GoogLeNet) · Cambridge Landmarks | median position error (m): 2.08 -> 1.30 | Improved |
| add LSTM-based structured dimensionality reduction (vs PoseNet) | PoseNet (GoogLeNet) · Cambridge Landmarks | median orientation error (deg): 6.83 -> 5.52 | Improved |
| add LSTM structured reduction instead of dropout (vs Bayesian PoseNet) | Bayesian PoseNet · Cambridge Landmarks | median position error (m): 1.92 -> 1.30 | Improved |
| add LSTM structured reduction (vs PoseNet) | PoseNet (GoogLeNet) · 7Scenes | median position error (m): 0.44 -> 0.31 | Improved |
| add LSTM structured reduction (vs PoseNet) | PoseNet (GoogLeNet) · 7Scenes | median orientation error (deg): 10.4 -> 9.85 | Improved |
| add LSTM structured reduction (vs PoseNet) | PoseNet (GoogLeNet) · 7Scenes - Heads | median orientation error (deg): 12.0 -> 13.7 | Hurt |
| add LSTM structured reduction (vs PoseNet) | PoseNet (GoogLeNet) · 7Scenes - Pumpkin | median position error (m): 0.61 -> 0.33 | Improved |
| add LSTM structured reduction (vs PoseNet) | PoseNet (GoogLeNet) · TUM-LSI | median position error (m): 1.87 -> 1.31 | Improved |
| add LSTM structured reduction (vs PoseNet) | PoseNet (GoogLeNet) · TUM-LSI | median orientation error (deg): 6.14 -> 2.79 | Improved |
| use CNN pose regression instead of SIFT-based Active Search | Active Search (w/) · Cambridge Landmarks | median position error (m): 0.36 -> 1.37 | Hurt |
| use CNN pose regression instead of SIFT-based Active Search | Active Search (w/) · 7Scenes | median position error (m): 0.06 -> 0.30 | Hurt |
| feed reshaped 32x64 matrix to four directional LSTMs instead of the raw 2048-D vector (ablation) | CNN+LSTM · Cambridge/7Scenes | pose accuracy | Improved |
Closest papers by meaning
- Learning Less Is More - 6D Camera Localization via 3D Surface Regression (CVPR 2018 · similarity 0.87)
- D2-Net: A Trainable CNN for Joint Description and Detection of Local Features (CVPR 2019 · similarity 0.86)
- Geometry-Aware Learning of Maps for Camera Localization (CVPR 2018 · similarity 0.85)
- From Coarse to Fine: Robust Hierarchical Localization at Large Scale (CVPR 2019 · similarity 0.85)
- HyperPose: Hypernetwork-Infused Camera Pose Localization and an Extended Cambridge Landmarks Dataset (CVPR 2025 · similarity 0.84)
- CAM Back Again: Large Kernel CNNs from a Weakly Supervised Object Localization Perspective (CVPR 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.
