Research map / Label-efficient and robust learning / Recognition and long-tail
A Hyperdimensional One Place Signature to Represent Them All: Stackable Descriptors For Visual Place Recognition
Connor Malone, Somayeh Hussaini, Tobias Fischer, Michael Milford
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 |
|---|---|---|---|
| fuse multiple reference sets with HDC bundling (HOPS) instead of best single reference | SALAD · Oxford RobotCar Dusk query | Recall@1: 76.3 -> 87.1 | Improved |
| fuse references with HOPS instead of best single reference | SALAD · Nordland (4 queries, avg) | Recall@1: absolute 2.9% | Improved |
| fuse references with HOPS | SALAD · SFU-Mountain Dusk query | Recall@1: 99.0% to 100% | Improved |
| use HOPS vs distance-matrix averaging and pooling | descriptors · all datasets | Recall@1: 69 out of 90 cases | Improved |
| use HOPS with 512D descriptors CosPlace/EigenPlaces on RobotCar Night query | CosPlace/EigenPlaces · Oxford RobotCar Night | Recall@1: 1.8% and 2.4% reduction | Hurt |
| use HOPS with CosPlace on Nordland Summer query | CosPlace · Nordland Summer | Recall@1: 1.0% reduction | Hurt |
| use pooling vs dMat averaging for low-dimensional descriptors | CosPlace/EigenPlaces · Oxford RobotCar | Recall@1: pooling better for low dimension | Mixed |
| reduce descriptor dimensionality with Gaussian random projection while fusing with HOPS (ablation) | CricaVPR · Oxford RobotCar Dusk | Recall@1: 10752D to 512D, 85.7 | No clear effect |
| replace real extra references with synthetic augmentations in HOPS | SALAD · Oxford RobotCar Dusk | Recall@1: 73.6 -> 76.1 | Improved |
| replace real extra references with synthetic augmentations in HOPS | SALAD · Oxford RobotCar Overcast | Recall@1: 84.8 -> 84.2 | Hurt |
| fuse 7-9 references per place with HOPS instead of keeping all references | SALAD · Google Landmarks v2 micro | Recall@1: 69.7 -> 65.8 | Hurt |
| use HOPS fusion instead of dimensionality reduction to 1024D at same memory | SALAD · Google Landmarks v2 micro | Recall@1: 59.7 -> 65.8 | Improved |
Closest papers by meaning
- Optimal Transport Aggregation for Visual Place Recognition (CVPR 2024 · similarity 0.88)
- Towards Implicit Aggregation: Robust Image Representation for Place Recognition in the Transformer Era (NeurIPS 2025 · similarity 0.87)
- SuperVLAD: Compact and Robust Image Descriptors for Visual Place Recognition (NeurIPS 2024 · similarity 0.87)
- BoQ: A Place is Worth a Bag of Learnable Queries (CVPR 2024 · similarity 0.85)
- Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition (AAAI 2025 · similarity 0.85)
- MutualVPR: A Mutual Learning Framework for Resolving Supervision Inconsistencies via Adaptive Clustering (NeurIPS 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.
