Research map / 3D vision / 3D generation
UniPR-3D: Towards Universal Visual Place Recognition with Visual Geometry Grounded Transformer
Tianchen Deng, Chen Xun, Ziming Li, Hongming Shen, Shuhao Zhai, Danwei Wang, Javier Civera, Hesheng Wang
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 3D VGGT tokens to 2D tokens in the descriptor (ablation) | UniPR-3D · Oxford2 sequence (2m) | R@1: 75.9 -> 81.4 | Improved |
| add 2D patch tokens to 3D-only descriptor (ablation) | UniPR-3D · MSLS val | R@1: 91.9 -> 94.1 | Improved |
| use optimal transport instead of GeM for 3D patch tokens (ablation) | UniPR-3D · Oxford2 sequence (2m) | R@1: 78.8 -> 81.4 | Improved |
| add explicit 3D pose injection (ablation) | UniPR-3D · Oxford2 sequence (2m) | R@1: 81.4 -> 78.3 | Hurt |
| add 2D cls and register tokens (ablation) | UniPR-3D · Oxford2 sequence (2m) | R@1: 80.3 -> 81.4 | Improved |
| increase test-time sequence length from 5 to 15 frames (trained on 5) (ablation) | UniPR-3D · Oxford2 sequence (2m) | R@1: 80.6 -> 92.4 | Improved |
| reduce test-time sequence length from 5 to 3 frames (ablation) | UniPR-3D · Oxford2 sequence (2m) | R@1: 80.6 -> 70.3 | Hurt |
| replace DINOv2 single-view descriptor with VGGT 3D-token descriptor (single-frame) | SALAD · single-frame VPR benchmarks | R@1: SF-XL v1 95.3 (MegaLoc) vs 93.8 UniPR-3D; Nordland 76.0 -> 78.4 | Mixed |
| use UniPR-3D instead of CaseVPR for sequence retrieval (strict 2m) | CaseVPR · Oxford2 sequence (2m) | R@1: 72.8 -> 80.6 | Improved |
| use VGGT backbone instead of DINOv2 backbone, latency | CaseVPR · Oxford1 | latency: 75 -> 140 | Hurt |
Closest papers by meaning
- Towards Implicit Aggregation: Robust Image Representation for Place Recognition in the Transformer Era (NeurIPS 2025 · similarity 0.91)
- Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition (AAAI 2025 · similarity 0.89)
- Optimal Transport Aggregation for Visual Place Recognition (CVPR 2024 · similarity 0.88)
- SuperVLAD: Compact and Robust Image Descriptors for Visual Place Recognition (NeurIPS 2024 · 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 0.87)
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.
