- Rating
- 3
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
The paper introduces UniPS, a training-free Visual Place Recognition (VPR) framework based on unified patch similarity. The method combines self-similarity-based patch weighting, weighted second-moment pooling for global retrieval, and patch-level similarity for reranking. The authors evaluate the approach on 17 standard and cross-environment benchmarks, reporting impressive results that outperform many trained and training-free baselines.
The extensive experimental evaluation and strong performance across diverse environments are clear strengths of the paper. However, my main concern is the limited methodological novelty, as the individual components are largely established techniques. Furthermore, several experimental comparisons require better control to establish the contribution of the proposed method.
Critical strengths
- The proposed training-free pipeline achieves impressive results across a broad range of VPR benchmarks, particularly in cross-environment settings.
- The unified framework is relatively simple and does not require additional training or labeled data.
- The extensive experiments, component ablations, and demonstrated applicability of the reranking module to existing methods provide useful empirical evidence.
Critical weaknesses
My main concern is that the proposed components, including self-similarity weighting, second-order pooling, and patch-based reranking, are already established techniques. The contribution therefore appears to lie primarily in their particular combination.
Moreover, the reranking formulation closely resembles EffoVPR (ICLR 2025), replacing attention-based patch selection with self-similarity weighting and MNN counting with weighted MaxSim, while even adopting the same threshold of . A more direct comparison would help clarify both the methodological novelty and the contribution of these modifications to performance.
Beyond the limited methodological novelty, it remains unclear how much of the reported performance improvement can be attributed to specific design choices rather than other factors. More controlled comparisons with closely related methods would help establish both the novelty and the actual contribution of the proposed components to the performance gains.
Questions for authors
Input resolution. Most baselines use input resolutions of 224 to 384 pixels, while UniPS uses 672 to 768 pixels. Beyond increasing runtime, higher resolution may also contribute to the performance gains. Please provide results at comparable resolutions.
Coarse ranking. The standalone performance of the proposed PCA-based global descriptor should be compared against CLS-token retrieval under matched settings and reported in Table 2. Reporting performance before reranking would help isolate the contribution of the proposed representation.
Feature extraction details. Appendix A.2 mentions value-facet features but does not specify the extraction layer. This is important since AnyLoc and EffoVPR-ZS use intermediate-layer features, whereas other models (e.g. SALAD, CricaVPR and more) use output-layer features. Please clarify the selected layer.
Missing top-K specification and ablation. The paper does not specify the value of K used for reranking, despite its impact on both retrieval accuracy and computational cost. Please clarify the value used and provide an ablation over different K values, reporting R@1 and runtime. Coarse-stage R@K should also be reported to distinguish candidate retrieval quality from reranking effectiveness.
Reranking generalisation. EffoVPR-ZS, the only other two-stage training-free baseline, is absent from this analysis. Given the similarity between the reranking mechanisms, applying the proposed reranker to the top-K candidates retrieved by EffoVPR-ZS would provide stronger evidence of its independent contribution.
Missing SoTA comparisons. Several relevant recent methods are absent from the quantitative comparisons, including EffoVPR, Pair-VPR (RA-L 2025), ImAge (NeurIPS 2025), SAGE (ICLR 2026), and SelaVPR++ (TPAMI 2026). Some report stronger results on individual standard VPR benchmarks. Please extend the comparison to these methods.
Minor Comments
- Line 192: CosPlace and EigenPlaces are not trained on the entire 41M-image SF-XL dataset. Please clarify the actual training set sizes.
- Line 202: What is the motivation or origin of the similarity template? Are the patch features L2-normalized so that their dot products correspond to cosine similarity?
- Figure 1: The bottom-right panel could benefit from visual refinement.
- Line 799: Missing period between sentences.
- Line 1125: "FUTUREWORK" should be "FUTURE WORK".
12 comments
Response to Reviewer hSsN (Part 2/5)
Ablation study on input resolution.
We also evaluated our method alone at different input resolutions (Table R3, with patch grids aligned between D2 and D3). Even at 322, OursD2 (79.3) and OursD3 (82.7) are above every trained method in Table R1 over the 17 benchmarks (the best is BoQ, 77.3). R@1 rises with resolution, peaking at about 672 for D2 and 640–768 for D3. Table R4 shows that at every resolution, K = 10 already captures most of the gain from reranking.
Table R3. R@1 vs. input resolution (grid-aligned, same tokens for D2 and D3).
| Model @ res | Tokens | Std (8) | Cross (9) | All (17) |
|---|---|---|---|---|
| OursD2 @224 | 256 | 84.3 | 70.1 | 76.8 |
| OursD2 @322 | 529 | 87.8 | 71.7 | 79.3 |
| OursD2 @336 | 576 | 88.2 | 74.2 | 80.8 |
| OursD2 @448 | 1024 | 89.2 | 75.3 | 81.8 |
| OursD2 @560 | 1600 | 90.1 | 75.9 | 82.6 |
| OursD2 @672 (Table 2) | 2304 | 90.3 | 76.3 | 82.9 |
| OursD2 @784 | 3136 | 90.3 | 76.0 | 82.7 |
| OursD3 @256 | 256 | 90.2 | 73.8 | 81.5 |
| OursD3 @322 | 400 | 91.2 | 75.2 | 82.7 |
| OursD3 @384 | 576 | 91.8 | 76.2 | 83.5 |
| OursD3 @512 | 1024 | 92.0 | 76.3 | 83.7 |
| OursD3 @640 | 1600 | 92.7 | 77.1 | 84.4 |
| OursD3 @768 (Table 2) | 2304 | 93.2 | 76.6 | 84.4 |
| OursD3 @896 | 3136 | 92.7 | 76.3 | 84.0 |
Table R4. Mean R@1 against shortlist size K at each input resolution; K=1 is the coarse ranking alone.
| Model @ res | Std K=1 | Std K=10 | Std K=1000 | Cross K=1 | Cross K=10 | Cross K=1000 |
|---|---|---|---|---|---|---|
| OursD2 @224 | 76.4 | 83.6 | 84.3 | 68.0 | 71.5 | 70.1 |
| OursD2 @322 | 76.4 | 85.4 | 87.8 | 63.3 | 72.1 | 71.7 |
| OursD2 @336 | 76.2 | 85.7 | 88.2 | 64.0 | 74.5 | 74.2 |
| OursD2 @448 | 77.4 | 86.8 | 89.2 | 65.2 | 73.6 | 75.3 |
| OursD2 @560 | 76.6 | 87.1 | 90.1 | 65.9 | 75.8 | 75.9 |
| OursD2 @672 (Table 2) | 81.2 | 87.4 | 90.3 | 65.9 | 75.2 | 76.3 |
| OursD3 @256 | 82.7 | 89.8 | 90.2 | 69.4 | 74.3 | 73.8 |
| OursD3 @322 | 83.3 | 89.9 | 91.2 | 67.6 | 75.4 | 75.2 |
| OursD3 @384 | 83.5 | 90.4 | 91.8 | 70.9 | 76.1 | 76.2 |
| OursD3 @512 | 82.1 | 90.2 | 92.0 | 70.8 | 76.6 | 76.3 |
| OursD3 @640 | 82.8 | 91.0 | 92.7 | 67.4 | 75.9 | 77.1 |
| OursD3 @768 (Table 2) | 82.5 | 91.5 | 93.2 | 70.2 | 75.6 | 76.6 |
| OursD3 @896 | 80.1 | 91.0 | 92.7 | 66.1 | 75.9 | 76.3 |
The full tables and analyses are included in Appendix B.4 (Tables 22 and 23).
Response to Reviewer hSsN (Part 3/5)
- The paper does not specify the value of K used for reranking. Please clarify the value used and provide an ablation over different K values, reporting R@1 and runtime. Coarse-stage R@K should also be reported to distinguish candidate retrieval quality from reranking effectiveness. (Q1 & Q4, W3)
All results of our method in the paper are reported at K = 1000, although K can be adjusted according to practical deployment requirements. We further conduct an ablation study on K, reporting coarse-ranking performance (R@K), reranked R@1, and the runtime of each stage using two backbones. As shown in Tables R5 and R6, K = 10 already achieves strong performance with very low coarse-ranking and reranking costs, demonstrating the effectiveness of coarse ranking alone. Specifically, for OursD3, coarse R@10 reaches 95.5 on standard and 93.2 on cross-environment benchmarks. Reranking only 10 candidates achieves a mean R@1 of 83.1 across the 17 benchmarks, just 1.3 points below K = 1000 (84.4), while reducing the total query time from 258.8 ms to 63.9 ms (Table R7). Increasing K to 100 further raises R@1 to 84.2, within 0.2 points of K = 1000. OursD2 exhibits the same trend.
Table R5. OursD3. Coarse = coarse-ranking R@K; R@1 = after reranking; Std (8), Cross (9), All (17).
| K | Coarse Std | Coarse Cross | R@1 Std | R@1 Cross | R@1 All |
|---|---|---|---|---|---|
| 1 | 82.5 | 70.2 | 82.5 | 70.2 | 76.0 |
| 10 | 95.5 | 93.2 | 91.5 | 75.6 | 83.1 |
| 50 | 98.4 | 99.2 | 92.7 | 76.6 | 84.2 |
| 100 | 98.9 | 99.5 | 92.8 | 76.6 | 84.2 |
| 200 | 99.4 | 99.7 | 93.1 | 76.6 | 84.4 |
| 500 | 99.7 | 99.8 | 93.2 | 76.6 | 84.4 |
| 1000 | 99.9 | 99.9 | 93.2 | 76.6 | 84.4 |
Table R6. OursD2. Same columns as Table R5.
| K | Coarse Std | Coarse Cross | R@1 Std | R@1 Cross | R@1 All |
|---|---|---|---|---|---|
| 1 | 81.2 | 65.9 | 81.2 | 65.9 | 73.1 |
| 10 | 92.2 | 91.3 | 87.4 | 75.2 | 80.9 |
| 50 | 96.1 | 98.3 | 89.1 | 76.5 | 82.4 |
| 100 | 97.3 | 98.8 | 89.5 | 76.3 | 82.5 |
| 200 | 98.2 | 99.3 | 89.9 | 76.3 | 82.7 |
| 500 | 99.3 | 99.6 | 90.2 | 76.3 | 82.8 |
| 1000 | 99.7 | 99.8 | 90.3 | 76.3 | 82.9 |
Table R7. Time (ms/query) by stage, mean over the 17 benchmarks (one H100).
| Method | K | Extraction | Coarse | Rerank | Total |
|---|---|---|---|---|---|
| OursD2 | 1 | 14.3 | 0.08 | – | 14.4 |
| 10 | 14.4 | 0.10 | 3.6 | 18.1 | |
| 50 | 14.4 | 0.10 | 20.3 | 34.8 | |
| 100 | 14.4 | 0.11 | 40.1 | 54.7 | |
| 1000 | 14.4 | 0.12 | 195.7 | 210.2 | |
| OursD3 | 1 | 60.9 | 0.10 | – | 61.0 |
| 10 | 60.1 | 0.10 | 3.6 | 63.9 | |
| 50 | 59.5 | 0.11 | 19.6 | 79.3 | |
| 100 | 59.5 | 0.15 | 41.5 | 101.3 | |
| 1000 | 58.7 | 0.15 | 199.9 | 258.8 |
In addition, we added a per-query, stage-wise timing comparison of UniPS-D2/D3 and other two-stage methods across all 17 benchmarks on one H100 (Table R8). Despite using a large backbone and high resolution, our method introduces little overhead beyond feature extraction. At K = 10, reranking takes only 3.6 ms, 6.6× faster than the fastest competing reranker (EffoVPR-ZS, 23.9 ms). OursD2 achieves the lowest total latency (18.1 ms) among all two-stage methods, while outperforming the best competitor, EffoVPR-ZS, by 6.8 R@1 points (80.9 vs. 74.1). Per-stage timings are provided in Appendix A.6 (Table 12).
Table R8. Per-query cost (ms), mean over the 17 benchmarks; Performance is the mean R@1 over the 17 benchmarks.
| Method | Extraction | Coarse | Reranking | Total | Performance |
|---|---|---|---|---|---|
| Patch-NetVLAD | 14.7 | 0.09 | 5189 | 5204 | 60.7 |
| R2Former | 3.9 | 0.05 | 63.5 | 67.4 | 62.6 |
| SelaVPR | 11.8 | 0.06 | 36.6 | 48.5 | 71.1 |
| EffoVPR-ZS | 6.2 | 0.06 | 23.9 | 30.2 | 74.1 |
| OursD2, K=10 | 14.4 | 0.10 | 3.6 | 18.1 | 80.9 |
| OursD3, K=10 | 60.1 | 0.10 | 3.6 | 63.9 | 83.1 |
| OursD2, K=100 | 14.4 | 0.11 | 40.1 | 54.7 | 82.5 |
| OursD3, K=100 | 59.5 | 0.15 | 41.5 | 101.3 | 84.2 |
| OursD2, K=1000 | 14.4 | 0.12 | 195.7 | 210.2 | 82.9 |
| OursD3, K=1000 | 58.7 | 0.15 | 199.9 | 258.8 | 84.4 |
Response to Reviewer hSsN (Part 1/5)
Thank you for your comments. We will address them as follows:
Similar to SegVLAD, our motivation for adopting high-resolution images is to fully utilise the visual information inherently available in the original images. Many of the commonly used datasets (e.g., Pitts, MSLS-val, St Lucia, Hawkins, Laurel Caverns, 17 Places, etc.) contain images with resolutions higher than 322 × 322, whereas many training-based baselines operate at substantially lower resolutions. For example, existing trained methods typically resize images to 224 × 224 during training (e.g., SALAD) to reduce the training time, and increase the resolution to 322 × 322 during inference (e.g., SALAD and BoQ). This creates a potential information bottleneck: visual details present in the original high-resolution images may already have been irreversibly lost during downsampling, potentially leading to retrieval failures.
Appendix C.3 (Figure 12) gives an example on Pitts30k (for the same query and its correct database image, the 640×640 input (40×40 patches) yields 115 patch matches and a reranking score of 12.8; at 256×256 (16×16 patches), 95 of these 115 matches disappear, the number of matches drops to 27 and the score falls to 8.5, below the 11.0 of an image from a wrong place, which is therefore ranked first. The lost matches lie mostly on fine details such as windows on the building facade, signs and traffic lights, which can no longer be resolved at low resolution). In contrast, our training-free approach is not constrained by the long training cost and can directly make full use of the visual information available at higher resolutions, avoiding the information loss from downsampling.
In addition, we conducted a systematic comparison across different input resolutions: all methods use the same DINOv3 backbone, and only the test-time input resolution changes (322 for all methods in Table R1, 768 in Table R2). As shown in Tables R1 and R2, our method achieves state-of-the-art performance at both resolutions. More specifically, at 322, OursD3 reaches a mean R@1 of 82.7 over all 17 benchmarks, 5.4 points above the best trained method, BoQ (77.3), and 11.7 points above it on the cross-environment benchmarks (75.2 vs. 63.5). At 768, OursD3 leads on both groups: +1.1 on the standard benchmarks (93.2 vs. 92.1), +10.0 on the cross-environment benchmarks (76.6 vs. 66.6) and +5.8 over all 17 (84.4 vs. 78.6).
Table R1. All methods on the same DINOv3 backbone at 322×322 input (R@1). † marks rows reported in Table 6.
Table R2. All methods on the same DINOv3 backbone at 768×768 input (R@1). † marks rows reported in Table 6.
The full tables and analyses are included in Appendix B.4 (Tables 20 and 21).