Research map / Detection and segmentation / Open-set and few-shot detection
Revisit Anything: Visual Place Recognition via Image Segment Retrieval
Kartik Garg, Sai Shubodh, Shishir N Y Kolathaya, Madhava Krishna, 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 |
|---|---|---|---|
| replace whole-image global aggregation (AnyLoc) with SuperSegment retrieval (SegVLAD-PreT) | AnyLoc DINOv2 · Baidu Mall | Recall@1/5: 75.2/87.6 -> 80.4/94.0 | Improved |
| replace whole-image global aggregation (AnyLoc) with SuperSegment retrieval on out-of-distribution InsideOut | AnyLoc DINOv2 · InsideOut | Recall@1/5: 2.4/8.0 -> 4.2/9.4 | Improved |
| replace SALAD global descriptor with SegVLAD-FineT on street-view data | SALAD · Pitts30K | Recall@1: 92.6 -> 93.1 | No clear effect |
| use segment retrieval on pretrained DINOv2 for Pitts30K | AnyLoc · Pitts30K | Recall@1/5: 87.7/94.7 -> 86.7/94.2 | Hurt |
| add neighborhood aggregation (SuperSegments) to segments (ablation) | SegVLAD · Baidu OOI instance retrieval | Recall@1: 64.1 -> 92.7 | Improved |
| expand segment neighborhood order from 0 to 3 with VLAD (ablation) | SegVLAD · Baidu Mall | Recall@1: 73.1 -> 77.7 | Improved |
| expand neighborhood with segment average pooling (ablation) | SAP · Baidu Mall | Recall@1: 74.6 -> 49.8 | Hurt |
| use VLAD aggregation instead of segment average pooling at order 3 (ablation) | SuperSegments · Baidu Mall | Recall@1: 49.8 -> 77.7 | Improved |
| use segment descriptors instead of random local point features (ablation) | SegVLAD · Baidu Mall | Recall@1: 69.1 -> 78.5 | Improved |
| replace SAM with FastSAM for segmentation (ablation) | SegVLAD · Baidu Mall | Recall@1: 78.5 -> 76.2 with 3.5s -> 0.28s segmentation | Mixed |
| filter database SuperSegments by IoU threshold 0.4 (ablation) | SegVLAD · AmsterTime | Recall@1: 58.9 -> 58.9 | No clear effect |
| use soft-assignment NetVLAD instead of hard VLAD (ablation) | SegVLAD · AmsterTime | Recall@1: 58.9 -> 60.2 | No clear effect |
Closest papers by meaning
- UniPR-3D: Towards Universal Visual Place Recognition with Visual Geometry Grounded Transformer (ECCV 2026 · similarity 0.90)
- Towards Implicit Aggregation: Robust Image Representation for Place Recognition in the Transformer Era (NeurIPS 2025 · similarity 0.89)
- Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition (AAAI 2025 · similarity 0.88)
- Optimal Transport Aggregation for Visual Place Recognition (CVPR 2024 · similarity 0.87)
- BoQ: A Place is Worth a Bag of Learnable Queries (CVPR 2024 · similarity 0.87)
- SuperVLAD: Compact and Robust Image Descriptors for Visual Place Recognition (NeurIPS 2024 · similarity 0.87)
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.
