Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 66 of the 66 references it cites were found in a registry or an index; none of the 64 sources its claims cite is checked for support yet.

Learning to Holistically Detect Bridges From Large-Size VHR Remote Sensing Imagery

Open access
Li, Yansheng, Luo, Junwei, Zhang, Yongjun, Tan, Yihua, Yu, Jin-Gang, Bai, Song · 2024 · IEEE Transactions on Pattern Analysis and Machine Intelligence

Abstract

Bridge detection in remote sensing images (RSIs) plays a crucial role in various applications, but it poses unique challenges compared to the detection of other objects. In RSIs, bridges exhibit considerable variations in terms of their spatial scales and aspect ratios. Therefore, to ensure the visibility and integrity of bridges, it is essential to perform holistic bridge detection in large-size very-high-resolution (VHR) RSIs. However, the lack of datasets with large-size VHR RSIs limits the deep learning algorithms' performance on bridge detection. Due to the limitation of GPU memory in tackling large-size images, deep learning-based object detection methods commonly adopt the cropping strategy, which inevitably results in label fragmentation and discontinuous prediction. To ameliorate the scarcity of datasets, this paper proposes a large-scale dataset named GLH-Bridge comprising 6,000 VHR RSIs sampled from diverse geographic locations across the globe. These images encompass a wide range of sizes, varying from 2,048 × 2,048 to 16,384 × 16,384 pixels, and collectively feature 59,737 bridges. These bridges span diverse backgrounds, and each of them has been manually annotated, using both an oriented bounding box (OBB) and a horizontal bounding box (HBB). Furthermore, we present an efficient network for holistic bridge detection (HBD-Net) in large-size RSIs. The HBD-Net presents a separate detector-based feature fusion (SDFF) architecture and is optimized via a shape-sensitive sample re-weighting (SSRW) strategy. The SDFF architecture performs inter-layer feature fusion (IFF) to incorporate multi-scale context in the dynamic image pyramid (DIP) of the large-size image, and the SSRW strategy is employed to ensure an equitable balance in the regression weight of bridges with various aspect ratios. Based on the proposed GLH-Bridge dataset, we establish a bridge detection benchmark including the OBB and HBB tasks, and validate the effectiveness of the proposed HBD-Net. Additionally, cross-dataset generalization experiments on two publicly available datasets illustrate the strong generalization capability of the GLH-Bridge dataset.

Overview

CoChat Summary

TL;DR

The paper introduces GLH-Bridge, a 6,000-image large-size VHR remote sensing dataset with 59,737 bridge annotations, and HBD-Net, a network for holistic bridge detection using separate detector-based feature fusion and shape-sensitive sample re-weighting. On the GLH-Bridge benchmark, HBD-Net achieves 35.35% mAP for OBB and 34.49% mAP for HBB, with strong AP on large/huge bridges and reported cross-dataset generalization.

Key findings

  • GLH-Bridge contains 6,000 VHR RSIs from over 400 cities/regions, with image sizes from 2,048 × 2,048 to 16,384 × 16,384 pixels, GSD 0.3–1.0 m, and 59,737 bridges manually annotated with both OBB and HBB; the split is 3,613 training, 1,194 validation, and 1,193 testing images.
  • HBD-Net achieves the best OBB result on GLH-Bridge with 35.35% mAP, 28.69% AP75, 33.47% AP_lg, and 20.61% AP_hg; for HBB it achieves 34.49% mAP, 35.21% AP_lg, and 35.59% AP_hg.
  • Ablations show SDFF without cross-scale-transfer distillation and IFF improves AP_hg by 8.57% over baseline; adding cross-scale-transfer distillation adds 3.4% AP_hg; IFF reaches 28.06% AP75 and 17.30% AP_hg. SSRW on the bottom-layer detector improves AP_lg by 1.37% and AP_hg by 3.30%, and the final HBD-Net improves AP75 by 2.40%, AP_lg by 4.17%, and AP_hg by 14.93% over baseline.
  • Cross-dataset generalization experiments report improvements of 3.79% on DOTA-v1.0 and 0.86% on DIOR-R; on DOTA-v1.0, HBD-Net reaches 46.11% mAP and 56.02% AP_BR.

Methods

  • Built GLH-Bridge from Google Earth and MapBox imagery collected 2019–2022, using NBI metadata and global city/river sampling; bridges were manually annotated with OBB and HBB under a three-stage annotation process with expert review.
  • Proposed HBD-Net with a separate detector-based feature fusion (SDFF) architecture over a dynamic image pyramid (DIP), inter-layer feature fusion (IFF), cross-scale-transfer distillation, and a shape-sensitive sample re-weighting (SSRW) strategy for aspect-ratio-sensitive regression weighting.
  • Established OBB and HBB benchmarks on GLH-Bridge against 18 object detection methods using MMRotate/MMDetection, with AP metrics including length-based AP_sh, AP_md, AP_lg, and AP_hg; experiments used one Tesla V100 GPU with 16 GB memory, batch size 1 for HBD-Net, downsampling ratio 2.0, and 1024×1024 cropping windows with 200-pixel overlap for comparison methods.

Limitations

  • The paper does not include a formal limitations section; it states future work will enrich GLH-Bridge in sample volume and sub-category annotation, and generalize HBD-Net to multi-class object detection in large-size images.
  • Evident from the text: GLH-Bridge is single-class (bridge only) and lacks sub-category labels, and the SSRW strategy is not used for the HBB task because horizontal-box aspect ratio does not accurately reflect whether bridges are elongated.
  • Evident from Table 6: zero-shot transfer from GLH-Bridge to DOTA-v1.0 and DIOR-R yields 15.78% and 15.01% AP, respectively, lower than in-domain training on those datasets (45.76% and 19.88%), indicating a generalization gap despite the paper’s reported relative improvements.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/tpami.2024.3393024

Cites

66 of 66 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

0 Valid · 5 Not valid · 42 Not checked · 128 Own results and interpretations

85 more claims

Create a free account to see all 175 claims and to verify the 63 references still open.

See all claims — free account
  • Probably not a substantial claim. 54
  • It may state two separate claims as one. 40
  • The quote only partly states this claim. 3

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 66 references

Cited by

See all 77

Metadata

Authors
Li, Y., Luo, J., Zhang, Y., Tan, Y., Yu, J.-G., & Bai, S.
Affiliations
  • Wuhan University (CN)
  • Huazhong University of Science and Technology (CN)
  • South China University of Technology (CN)
  • ByteDance (China) (CN)
Venue
IEEE Transactions on Pattern Analysis and Machine Intelligence
Published
2024-12
Volume
46
Issue
12
Pages
11507-11523
ISSN
  • 0162-8828 (ISSN-L)
  • 2160-9292
  • 1939-3539
OpenAlex
W4396241435
Cited by
77
References
66
Citations by year
2026: 34 · 2025: 34 · 2024: 9
FWCI
6.65
Type
journal article
Language
en
Open access
Locations
  • IEEE Transactions on Pattern Analysis and Machine Intelligence (publishedVersion)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
National Natural Science Foundation of China (42371321, 42030102)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Automated Road and Building Extraction
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Remote-Sensing Image Classification
Keywords
bridge detection, oriented bounding boxes, feature fusion, aspect ratio, sample reweighting, object detection, cross-dataset generalization, infrastructure monitoring

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-19, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-19 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Learning to Holistically Detect Bridges From Large-Size VHR Remote Sensing Imagery. https://app.cochat.ai/papers/10.1109/tpami.2024.3393024
The paper
Li, Y., Luo, J., Zhang, Y., Tan, Y., Yu, J.-G., & Bai, S. (2024). Learning to Holistically Detect Bridges From Large-Size VHR Remote Sensing Imagery. IEEE Transactions on Pattern Analysis and Machine Intelligence. https://doi.org/10.1109/tpami.2024.3393024

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.