Research map / Label-efficient and robust learning / Recognition and long-tail
VLAD-BuFF: Burst-aware Fast Feature Aggregation for Visual Place Recognition
Ahmad Khaliq, Ming Xu, Stephen Hausler, Michael J Milford, Sourav Garg
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 burstiness-discounted (soft count) weighting to VLAD aggregation (ablation) | NetVLAD, P30K-Trip-R18 · MSLS (val) | Recall@1: 46.8 -> 50.0 | Improved |
| add burstiness weighting, DINOv2 backbone with Pitts30k training (ablation) | NetVLAD · MSLS (val) | Recall@1: 71.3 -> 72.8 | Improved |
| add burstiness weighting with large SFXL-trained CosPlace-style models (ablation) | VLAD · MSLS (val) | Recall@1: 80.1 -> 80.3 | No clear effect |
| add burstiness weighting to GSV-MS-DINOv2 VLAD (8K) (ablation) | VLAD · MSLS (val) | Recall@1: 91.8 -> 92.4 | Improved |
| learn local features jointly with burstiness weights instead of post-processing (No-Train Weight) (ablation) | VLAD-BuFF · Tokyo247 | Recall@1: 57.46 -> 59.37 | Improved |
| replace burstiness weighting with learned PixelMLP saliency weights (ablation) | VLAD-BuFF · Tokyo247 | Recall@1: 59.37 -> 57.14 | Hurt |
| use PCA-initialized pre-pool projection instead of random-init linear or MLP (ablation) | VLAD-BuFF · MSLS (val) | Recall@1 | Improved |
| use randomly-initialized MLP for pre-pool projection (ablation) | VLAD-BuFF · MSLS (val) | Recall@1 | Hurt |
| add pre-pool projection 768 to 192 dims (ablation) | VLAD-BuFF · aggregation time | time (ms): 44.8 -> 14.2 | Improved |
| burstiness weighting vs vanilla VLAD with several pre-pool dimensions (ablation) | VLAD-BuFF · MSLS (val) | Recall@1: 90.4 -> 91.6 (256-dim); 92.0 -> 92.6 (384-dim) | Improved |
Closest papers by meaning
- SuperVLAD: Compact and Robust Image Descriptors for Visual Place Recognition (NeurIPS 2024 · similarity 0.91)
- Towards Implicit Aggregation: Robust Image Representation for Place Recognition in the Transformer Era (NeurIPS 2025 · similarity 0.91)
- Optimal Transport Aggregation for Visual Place Recognition (CVPR 2024 · similarity 0.90)
- Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition (AAAI 2025 · similarity 0.88)
- MutualVPR: A Mutual Learning Framework for Resolving Supervision Inconsistencies via Adaptive Clustering (NeurIPS 2025 · similarity 0.88)
- Towards Seamless Adaptation of Pre-trained Models for Visual Place Recognition (ICLR 2024 · 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.
