Organizations: School of Artificial Intelligence, Beijing University of Posts and Telecommunications · Key Laboratory of Computing Power Network and Information Security, Ministry of Education, Shandong Computer Science Center, Qilu University of Technology
Visual Place Recognition (VPR) localizes a query image by retrieving database images of the same or nearby place, yet its robustness is often degraded by domain shifts arising from illumination, weather, seasonal changes, and dynamic occlusions. One contributing factor is the limited appearance diversity of the same place in existing training data. To address this issue, we propose AdaptVPR, a route-aware generative augmentation framework that constructs same-place hard positives for robust VPR training. AdaptVPR first uses a vision language model to parse scene attributes and estimate editing feasibility, while a rule-based scheduler determines the generation route according to editability scores and risk constraints. The generation process is decomposed into three complementary routes: the Global Appearance Route introduces global scene changes in weather, illumination, and time of day; the Local Occlusion Route inserts plausible dynamic occluders; and the Dual Route combines both types of perturbations to produce more challenging appearance shifts. Each generated candidate is evaluated using a VPR-oriented verification scheme based on geometric consistency and appearance diversity, reducing the risk of structural drift while ensuring sufficient appearance variation. Global candidates are generated once and rejected if verification fails, while Local Occlusion and Dual candidates use verification feedback for limited prompt refinement and regeneration. Using this framework, we construct AdaptCities, containing 160K verified synthetic same-place hard positives. Experiments across multiple VPR baselines and vision foundation backbones show consistent gains on standard benchmarks and substantial improvements under challenging domain shifts, with R@1 gains of up to 9.2%. The source code and data resources are publicly available at https://github.com/chenshunpeng/AdaptVPR.
Figures & tables
Fig. 1: VPR-oriented hard positive generation. Given the same reference image and seed instruction, generic editing methods may introduce structure drift, identity drift, or uncontrolled object insertion. AdaptVPR decomposes generation into the Global Appearance Route, Local Occlusion Route, and Dual Route, and uses VPR-oriented verification to retain visually diverse yet geometrically consistent same-place positives. For the Local Occlusion Route and Dual Route, verification feedback further guides prompt refinement through a reflective generation loop.
Fig. 2: R@1 gains from AdaptVPR across representative VPR methods.
Fig. 3: Overview of AdaptVPR. A VLM first parses each reference image and estimates its scene attributes and editability, after which a rule-based scheduler selects the Global Appearance, Local Occlusion, Dual, or Skip Route according to the parsed scene state and routing rules. The corresponding generators then synthesize candidate hard positives, which are evaluated for geometric consistency and appearance diversity. The Global Appearance Route performs a single generation followed by verification, whereas failed Local Occlusion and Dual candidates can be refined using verification feedback. Candidates satisfying the geometry and diversity gate are retained to construct AdaptCities for downstream VPR training.
Route / Subtype
# Images
Share (%)
Total
160,000
100.0
Global Appearance Route
50,239
31.4
Overcast
29,607
18.5
Snowy
9,367
5.9
Nighttime
5,907
3.7
Other global conditions
5,358
3.3
TABLE I: Composition of AdaptCities by generation route and subtype.
ID
Dataset
Primary Variation
Database
Queries
1
Pitts30k-test
Viewpoint
10,000
6,816
2
MSLS-val
Mixed conditions
18,871
740
3
Tokyo24/7
Day–night
75,984
315
4
Nordland
Seasonal
27,592
27,592
5
SVOX
Weather / illumination
17,166
14,278
6
Pitts250k-test
Large-scale urban
83,952
8,280
TABLE II: Statistics and dominant visual variations of the benchmark datasets used for evaluation. The three SF-XL query sets share the same large-scale database.
TABLE III: Main comparison with representative VPR methods on multiple benchmark datasets. All models are evaluated at a resolution of 322×322 . Baseline results are taken from the original papers when available; otherwise, we evaluate the officially released pretrained models under the corresponding benchmark protocols. “+AdaptVPR” denotes training the corresponding method with our generated hard positives while keeping its model configuration unchanged. Red and green values in parentheses denote the absolute R@1 gains and drops relative to the corresponding baseline, respectively.
TABLE IV: Comparison under challenging domain conditions including large-scale retrieval, cross-season changes, occlusion, and nighttime appearance variations.
Baseline
Filtering
Retained Ratio
Training Time (min/epoch)
Pitts30k
MSLS-val
SF-XL-Night
SF-XL-Occ.
SF-XL-v1
Geo.
Div.
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
BoQ
-
-
100.0%
14.6
93.4
96.7
97.7
93.1
96.4
96.9
49.4
62.2
66.1
50.0
64.5
68.4
90.7
94.3
95.4
✓
-
81.7%
13.1
93.3
96.9
97.8
93.2
95.9
96.6
49.1
60.1
66.5
48.7
63.2
67.1
89.2
93.9
95.2
-
✓
83.7%
13.9
93.2
96.8
97.8
93.2
96.4
97.0
46.8
59.9
64.8
48.7
64.5
69.7
88.6
93.3
94.5
✓
✓
65.6%
13.8
93.8
97.4
98.3
94.3
97.0
97.8
50.0
64.2
68.5
52.6
68.4
72.4
91.5
95.1
95.8
TABLE V: Ablation study of synthetic quality filtering in AdaptVPR. We evaluate the effect of geometry consistency and appearance diversity for filtering generated hard-positive samples. “Retained Ratio” denotes the percentage of candidates retained from the generated candidate pool after filtering. Training time is averaged over 10 epochs on a single NVIDIA RTX PRO 6000 GPU. The 1st , 2nd , and 3rd -best values are highlighted.
Baseline
Route
Training Time (min/epoch)
Pitts30k
MSLS-val
SF-XL-Night
SF-XL-Occ.
SF-XL-v1
Global
Local
Dual
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
BoQ
Unfiltered
14.6
93.4
96.7
97.7
93.1
96.4
96.9
49.4
62.2
66.1
50.0
64.5
68.4
90.7
94.3
95.4
✓
-
-
12.0
93.1
96.8
97.7
93.0
96.1
96.8
48.1
59.2
62.9
50.0
63.2
69.7
89.3
93.1
94.5
-
✓
-
12.0
93.3
96.8
97.7
92.8
96.2
96.9
49.6
60.5
65.2
52.6
64.5
72.4
89.1
93.7
95.2
-
-
✓
12.6
93.3
96.8
97.8
93.2
96.4
96.8
47.4
59.9
64.6
51.3
63.2
67.1
89.8
93.8
95.1
✓
✓
✓
13.8
93.8
97.4
98.3
94.3
97.0
97.8
50.0
64.2
68.5
52.6
68.4
72.4
91.5
95.1
95.8
TABLE VI: Ablation study of different generation routes in AdaptVPR on challenging VPR benchmarks. The first row uses all unfiltered synthetic samples. Starting from the BoQ baseline, we evaluate the contribution of the Global Appearance Route, Local Occlusion Route, and Dual Route. Training time is averaged across epochs on a single NVIDIA RTX PRO 6000 GPU.
Fig. 6: Training dynamics of BoQ with and without AdaptVPR over the first 20 epochs. AdaptVPR produces higher training loss, reflecting increased difficulty from generated same-place hard positives, while optimization remains stable and Pitts30k-val R@1 stays higher during most of training.
Fig. 7: Qualitative examples of AdaptVPR under three generation routes. (a) The Global Appearance Route introduces scene-level changes in weather, illumination, and time of day through one-shot generation while preserving place structure. (b) The Local Occlusion Route introduces dynamic foreground occlusions and refines failed candidates using verification feedback. (c) The Dual Route combines global appearance changes with local occlusions and similarly refines failed candidates through verification-guided prompt updates. The final accepted samples exhibit challenging visual variations while preserving key structural cues of the original place, providing verified hard positives for VPR training.
Generation Strategy
Route Selection
Verifier Feedback
Structured Prompt
Avg. Attempts
Avg. Time (s)
AR (%)
One-shot
Random
✗
✗
1.00
21.0
50.8
AdaptVPR
Adaptive
✓
✓
2.24
46.5
65.6
TABLE VII: Analysis of the complete generation procedure in AdaptVPR. We compare a one-shot baseline with the full AdaptVPR procedure, which combines adaptive route selection, structured prompts, verifier feedback, and repeated generation within the reflection budget. We report the average number of attempts, end-to-end generation time, and candidate acceptance rate (AR).
Method
Backbone
Params. (M)
Train Res.
Test Res.
Pitts30k
MSLS-val
Tokyo24/7
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
SALAD
DINOv2-B
88.0
224 × 224
322 × 322
92.5
96.4
97.5
92.2
96.4
97.0
94.6
97.5
97.8
+AdaptVPR
DINOv2-B
88.0
224 × 224
322 × 322
93.2 (+0.7)
96.8
97.8
93.5 (+1.3)
96.8
97.2
94.9 (+0.3)
98.1
98.4
SALAD
DINOv3-B
87.1
224 × 224
320 × 320
92.7
96.5
97.6
91.1
95.7
96.6
98.4
98.7
99.1
+AdaptVPR
DINOv3-B
87.1
224 × 224
320 × 320
93.2 (+0.5)
96.9
97.7
92.4 (+1.3)
97.2
98.0
98.7 (+0.3)
98.7
99.4
TABLE VIII: Generalization of AdaptVPR across vision foundation backbones. We compare DINOv2-B and DINOv3-B under their corresponding training and inference resolutions on Pitts30k, MSLS-val, and Tokyo24/7.
Real:Synthetic Batch Ratio
SF-XL-v1
SF-XL-Night
SF-XL-Occ.
Pitts30k
MSLS-val
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
R@1
R@5
R@10
Real-only
89.2
93.1
94.9
47.6
59.7
64.4
51.3
63.2
68.4
93.0
96.7
97.7
93.2
96.1
96.6
8:1
91.5
95.1
95.8
50.0
64.2
68.5
52.6
68.4
72.4
93.8
97.4
98.3
94.3
97.0
97.8
4:1
89.9
94.6
94.7
49.4
60.3
65.9
50.7
65.8
70.4
93.2
96.8
97.7
93.4
97.0
97.4
2:1
88.6
94.3
95.2
48.7
59.4
65.2
50.0
61.8
67.1
93.1
96.8
97.8
92.6
95.9
96.4
1:1
88.4
94.3
95.3
48.9
59.7
64.6
51.3
61.8
67.1
93.1
96.7
97.7
92.7
95.9
96.5
TABLE IX: Effect of the real-to-synthetic sampling ratio in AdaptVPR. The full verified synthetic pool is available in all mixed settings, while only the batch composition is varied. All settings use the same number of optimization steps.
Visual place recognition (VPR) aiming at predicting the location of an image based solely on its visual features is a fundamental task in robotics and autonomous systems. Domain variation remains one of the main challenges in VPR and is relatively unexplored. Existing VPR models attempt to achieve domain agnosticism either by training on large-scale datasets that inherently contain some domain variations, or by being specifically adapted to particular target domains. In practice, the former lacks explicit domain supervision, while the latter generalizes poorly to unseen domain shifts. This paper proposes a novel query-based domain-agnostic VPR model called QdaVPR. First, a dual-level adversarial learning framework is designed to encourage domain invariance for both the query features forming the global descriptor and the image features from which these query features are derived. Then, a triplet supervision based on query combinations is designed to enhance the discriminative power of the global descriptors. To support the learning process, we augment a large-scale VPR dataset using style transfer methods, generating various synthetic domains with corresponding domain labels as auxiliary supervision. Extensive experiments show that QdaVPR achieves state-of-the-art performance on multiple VPR benchmarks with significant domain variations. Specifically, it attains the best Recall@1 and Recall@10 on nearly all test scenarios: 93.5%/98.6% on Nordland (seasonal changes), 97.5%/99.0% on Tokyo24/7 (day-night transitions), and the highest Recall@1 across almost all weather conditions on the SVOX dataset. Our code will be released at https://github.com/shuimushan/QdaVPR.
Shanshan Wan, Lai Kang, Yingmei Wei +3
College of Systems Engineering and the Laboratory for Big Data and Decision, National University of Defense Technology, Changsha, Hunan 410073, China
Visual Place Recognition (VPR) aims to determine the geographic location of a query image by retrieving its most visually similar counterpart from a geo-tagged reference database. Recently, the emergence of the powerful visual foundation model, DINOv2, trained in a self-supervised manner on massive datasets, has significantly improved VPR performance. This improvement stems from DINOv2's exceptional feature generalization capabilities but is often accompanied by increased model complexity and computational overhead that impede deployment on resource-constrained devices. To address this challenge, we propose D2-VPR, a Distillation- and Deformable-based framework that retains the strong feature extraction capabilities of visual foundation models while significantly reducing model parameters and achieving a more favorable performance-efficiency trade-off. Specifically, first, we employ a two-stage training strategy that integrates knowledge distillation and fine-tuning. Additionally, we introduce a Distillation Recovery Module (DRM) to better align the feature spaces between the teacher and student models, thereby minimizing knowledge transfer losses to the greatest extent possible. Second, we design a Top-Down-attention-based Deformable Aggregator (TDDA) that leverages global semantic features to dynamically and adaptively adjust the Regions of Interest (ROI) used for aggregation, thereby improving adaptability to irregular structures. Extensive experiments demonstrate that our method achieves competitive performance compared to state-of-the-art approaches. Meanwhile, it reduces the parameter count by approximately 64.2% (compared to CricaVPR).Code is available at https://github.com/tony19980810/D2VPR.
Zheyuan Zhang, Jiwei Zhang, Boyu Zhou +2
School of Computer Science, Beijing University of Posts and Telecommunications · Key Laboratory of Interactive Technology and Experience System, Ministry of Culture and Tourism, Beijing University of Posts and Telecommunications
Visual Place Recognition (VPR) determines a query image's geographic location by matching it against geotagged databases. However, existing methods struggle with perceptual aliasing caused by irrelevant regions and inefficient re-ranking due to rigid candidate scheduling. To address these issues, we introduce FoL++, a method combining robust discriminative region modeling with adaptive re-ranking. Specifically, we propose a Reliability Estimation Branch to generate spatial reliability maps that explicitly model occlusion resistance. This representation is further optimized by two spatial alignment losses (SAL and SCEL) to effectively align features and highlight salient regions. For weakly supervised learning without manual annotations, a pseudo-correspondence strategy generates dense local feature supervision directly from aggregation clusters. Our Adaptive Candidate Scheduler dynamically resizes candidate pools based on global similarity. By weighting local matches by reliability and adaptively fusing global and local evidence, FoL++ surpasses traditional independent matching systems. Extensive experiments across seven benchmarks demonstrate that FoL++ achieves state-of-the-art performance with a lightweight memory footprint, improving inference speed by 40% over FoL. Code and models will be released (and merged with FoL) at https://github.com/chenshunpeng/FoL.
Shunpeng Chen, Yukun Song, Changwei Wang +6
School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing 100876, China. · Key Laboratory of Computing Power Network and Information Security, Ministry of Education, Shandong Computer Science Center, Qilu University of Technology. · Spatialtemporal AI. +1