Pith. sign in

REVIEW 2 major objections 5 minor 54 references

Patch-level alignment plus dual hypergraphs let RGB-thermal video detectors handle real misalignment and set new accuracy marks on a 1,000-video drone benchmark.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-10 11:35 UTC pith:RWYUDVQE

load-bearing objection Useful RGBT video detector plus a genuinely large drone benchmark; the SOTA numbers are real but partly hard to interpret because GT is RGB-only and residual alignment is never measured. the 2 major comments →

arxiv 2607.08191 v1 pith:RWYUDVQE submitted 2026-07-09 cs.CV

Dual-Correlation Hypergraph Network for Unaligned RGBT Video Object Detection and A Large-scale Benchmark

Qishun Wang , Yapeng Li , Bin Luo , Zhengzheng Tu , Chenglong Li This is my paper
classification cs.CV
keywords RGB-Thermalvideo object detectionmultimodal fusionhypergraphspatial misalignmentbenchmark datasetdrone vision
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

RGB-thermal video object detection fails in practice when the two cameras are not perfectly registered and when the only public video benchmark is small and traffic-only. This paper claims that treating misalignment as a local, region-varying problem and then fusing the corrected streams with hypergraphs that capture both temporal and cross-modal higher-order links solves both issues. The method first warps thermal patches onto RGB with predicted affine transforms and injects RGB edge structure into thermal; it then builds two hypergraphs—one across time, one across modalities—to aggregate complementary object evidence. The authors also release DVT-VOD1000, a drone-captured set of 1,000 sequences and more than 100,000 pairs spanning campuses, night, rain and snow. On that set and the earlier VT-VOD50 the resulting network records the highest reported AP50 scores among compared detectors.

Core claim

After correcting regionally varying RGB-thermal misalignment with patch-wise affine transforms and a global LBP injection, dual hypergraphs that separately model temporal and multimodal higher-order correlations produce state-of-the-art video object detection accuracy on both the existing VT-VOD50 benchmark and the new large-scale DVT-VOD1000 dataset.

What carries the argument

Dual-Correlation Hypergraph Network (DHNet): Patch-based Spatial Alignment Module (PSAM) that predicts a 6-parameter affine transform per non-overlapping patch, followed by Dual Hypergraph Fusion Module (DHFM) that builds two sparse hypergraphs (temporal and multimodal) whose message-passing aggregates complementary object features.

Load-bearing premise

Independent per-patch affine warps plus one global edge-map injection are enough to correct the real misalignment that remains after the drone platform’s own registration, and a fixed distance threshold of 8 reliably selects the hyperedges that carry the useful complementary information.

What would settle it

Retrain the same backbone with PSAM and DHFM disabled (or with a single global affine instead of per-patch warps) on DVT-VOD1000 and check whether AP50 still exceeds the strongest published multimodal image detector on that split; a drop that erases the claimed margin would falsify the necessity of the dual-correlation design.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The manuscript proposes DHNet for unaligned RGB-Thermal video object detection. It first applies a Patch-based Spatial Alignment Module (PSAM) that predicts independent affine transforms (Eq. 3) on non-overlapping patches of multimodal (and temporal) features, then injects global LBP positional cues from RGB into thermal. A Dual Hypergraph Fusion Module (DHFM) subsequently builds separate temporal and cross-modal hypergraphs using Euclidean-distance thresholding (τ = 8) to aggregate higher-order complementary information before the YOLOV8 detection head. The authors also release DVT-VOD1000, a drone-captured RGBT VOD benchmark of 1,000 sequences / >103k pairs spanning 15 categories and diverse scenes (day/night, rain, snow, campus, rural, traffic). Experiments on VT-VOD50 and DVT-VOD1000 report SOTA AP50 (57.5 % / 31.7 % for DHNet-L) against a broad set of image- and video-based detectors, with ablations isolating PSAM, LBP and DHFM contributions.

Significance. If the gains hold under residual-misalignment scrutiny, the work supplies both a practical architecture for a realistic (weakly aligned) sensing regime and a substantially larger, multi-scene drone RGBT VOD benchmark that the community currently lacks. The public release of code and DVT-VOD1000 is a concrete contribution. Ablations (Tables IV–VI) and comparisons against many recent detectors give a reasonably complete empirical picture. The central technical claim—that local affine patches plus hypergraph dual-correlation extract genuine cross-modal complementarity rather than merely re-weighting RGB features—remains only partially stress-tested because all ground-truth boxes are RGB-only.

major comments (2)
  1. Section IV-C states that all annotations are performed exclusively on the RGB modality “to prevent potential conflicts arising from spatial misalignment.” Consequently every AP number in Tables II–III measures localization success against RGB boxes only. After the fixed 5×5-patch affine (Eq. 3) + single global LBP injection, residual misalignment is never quantified (no mean endpoint error of object centroids, no thermal re-annotation of a subset). Without such a residual-alignment metric or a thermal-box re-evaluation experiment, it is difficult to separate true dual-correlation gains from ordinary multi-scale RGB enhancement or soft attention supplied by the thermal stream. This is load-bearing for the paper’s central claim.
  2. Section III-C and the free-parameter list: the hyperedge construction relies on a hard Euclidean threshold τ = 8 that is never ablated or justified beyond “set to 8 in this work.” Table V varies only PSAM patch size; no sensitivity study of τ (or of alternative similarity measures) is provided. Because the reported AP improvements of DHFM over Cross-Attention / DCEvo (Table VI) rest on these hyperedges, a short sensitivity analysis is required to establish that the gains are not an artifact of a single tuned threshold.
minor comments (5)
  1. Abstract and Section IV claim “1,000 video sequences with 103,464 RGBT image pairs,” yet Section IV-D and Table I state “500 RGBT video pairs” / 206,928 frames. Clarify whether 1,000 counts RGB+thermal streams separately or whether the numbers are inconsistent.
  2. Figure 1 caption and surrounding text assert “weak spatial alignment” but supply no quantitative misalignment statistics (pixel shift histograms, etc.) for either VT-VOD50 or DVT-VOD1000.
  3. Table II lists FPS for DHNet-S as 73 and for DHNet-L as 23.3; the same table reports YOLOV13-S at 476 FPS. Hardware and batch-size conditions for the FPS column should be stated uniformly.
  4. Notation: f^RGB_t versus F^RGB_t is used interchangeably for features at different stages; a short glossary or consistent casing would help.
  5. Related-work coverage of recent RGBT alignment methods (beyond RGFNet) and of hypergraph object detectors is thin; a few additional citations would situate DHFM more clearly.

Circularity Check

0 steps flagged

Empirical CV paper with independent modules, new dataset, and held-out SOTA numbers; only minor non-load-bearing self-citations to prior baselines.

full rationale

DHNet is an end-to-end trained detector (YOLOv8 backbone + PSAM local affine patches + LBP injection + DHFM hypergraphs with fixed τ=8). All reported AP50/AP gains (Tables II–III) and ablations (Tables IV–VI) are measured on held-out splits of VT-VOD50 and the newly collected DVT-VOD1000 against external and prior baselines. No equation reduces a claimed prediction to a fitted constant by construction; the affine matrices (Eq. 3) and hyperedge matrix H are learned/thresholded from features, not defined from the target metric. Self-citations to the authors’ EINet and PTMNet appear only as comparison methods and task introduction, not as uniqueness theorems or ansatze that force the architecture or numbers. The RGB-only annotation choice (Sec. IV-C) is a methodological decision that may affect interpretation of complementarity but does not create definitional circularity in the derivation chain. The paper is therefore self-contained against external benchmarks; score remains in the 0–2 band.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 3 invented entities

The central empirical claims rest on standard deep-learning practice plus a handful of hand-chosen architectural constants and the modeling assumption that local affine warps plus LBP suffice for the observed misalignment. No new physical entities are postulated; the free parameters are ordinary hyper-parameters of the network and fusion modules.

free parameters (3)
  • hyperedge distance threshold τ = 8
    Fixed to 8 to sparsify the hypergraph; chosen by the authors and not derived from first principles.
  • PSAM patch size = 5×5
    Selected as 5×5 after a three-point ablation; controls the granularity of local affine alignment.
  • network depth/width scaling (DHNet-S vs DHNet-L)
    Standard YOLO-style scaling factors chosen to trade accuracy for speed; not predicted by theory.
axioms (4)
  • domain assumption RGB and thermal modalities supply complementary object cues that remain useful after imperfect spatial alignment.
    Stated throughout the introduction and method; underpins the entire fusion design.
  • ad hoc to paper Local affine transforms predicted from concatenated patch descriptors can correct regionally varying sensor misalignment sufficiently for detection.
    Core modeling choice of PSAM (Section III-B, Eqs. 1–3).
  • ad hoc to paper Euclidean distance in feature space with a hard threshold yields meaningful higher-order hyperedges for both temporal and cross-modal aggregation.
    Construction of the incidence matrix H inside DHFM (Section III-C).
  • domain assumption Standard YOLO detection loss and training recipe remain appropriate after the added alignment and hypergraph modules.
    Explicitly adopted without modification (Section III-A).
invented entities (3)
  • Patch-based Spatial Alignment Module (PSAM) no independent evidence
    purpose: Perform decentralized local affine warps to reduce regionally varying RGB–thermal misalignment before fusion.
    New architectural block introduced in Section III-B; no independent physical existence outside the network.
  • Dual Hypergraph Fusion Module (DHFM) no independent evidence
    purpose: Jointly model higher-order temporal and cross-modal correlations via two hypergraphs.
    New fusion block of Section III-C; builds on prior hypergraph detectors but is specific to this dual-correlation design.
  • DVT-VOD1000 dataset independent evidence
    purpose: Provide a large-scale, scene-diverse, deliberately misaligned RGBT video benchmark for evaluation.
    Newly collected and annotated resource described in Section IV; will be released publicly.

pith-pipeline@v1.1.0-grok45 · 20715 in / 2835 out tokens · 31906 ms · 2026-07-10T11:35:56.396303+00:00 · methodology

0 comments
read the original abstract

RGB-Thermal (RGBT) Video Object Detection (VOD) has gained significant traction due to its ability to overcome the limitations of conventional RGB-based VOD under challenging conditions. However, spatial misalignment commonly exists between RGBT image pairs. To address this, we propose a Dual-Correlation Hypergraph Network (DHNet) that captures high-dimensional complementary information by explicitly modeling two types of correlations: temporal correlation across consecutive frames and spatial correlation from cross-modal features. Specifically, we first design a Patch-based Spatial Alignment Module (PSAM) to sequentially align the multimodal features at the local region level. Subsequently, we introduce a Dual Hypergraph Fusion Module (DHFM), which constructs separate temporal and multimodal hypergraphs to enhance object discriminability through dual-correlation learning. Furthermore, the field currently lacks a large-scale, scene-diverse benchmark dataset for comprehensive evaluation. To address this gap, we construct DVT-VOD1000, a large-scale RGBT VOD dataset containing 1,000 video sequences with 103,464 RGBT image pairs. The dataset covers diverse scenarios, including campuses, parks, transportation, rural areas, night scenes, rain, and snow. Comprehensive experiments on VT-VOD50 and our DVT-VOD1000 demonstrate that DHNet achieves state-of-the-art detection accuracy. The dataset and source code will be made publicly available on https://github.com/tzz-ahu/ to support academic research.

Figures

Figures reproduced from arXiv: 2607.08191 by Bin Luo, Chenglong Li, Qishun Wang, Yapeng Li, Zhengzheng Tu.

Figure 1
Figure 1. Figure 1: Existing VT-VOD50 and our proposed DVT-VOD1000, commonly [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The core architecture of DHNet. The PSAM first performs local spatial alignment on the two input feature maps in a region-by-region manner. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the patch-based alignment process in the PSAM. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The DVT-VOD1000 dataset comprises a diverse range of RGBT image pairs captured across various typical scenarios. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a) depicts the video sequences in the DVT-VOD1000 dataset and their length distribution. (b) presents the distribution of the 15 object categories. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages · 5 internal anchors

  1. [1]

    Highly accurate moving object detection in variable bit rate video-based traffic monitoring systems,

    S.-C. Huang and B.-H. Chen, “Highly accurate moving object detection in variable bit rate video-based traffic monitoring systems,”IEEE transactions on neural networks and learning systems, vol. 24, no. 12, pp. 1920–1931, 2013

    work page 1920

  2. [2]

    Anomaly detection in auto- mobile control network data with long short-term memory networks,

    A. Taylor, S. Leblanc, and N. Japkowicz, “Anomaly detection in auto- mobile control network data with long short-term memory networks,” in2016 IEEE international conference on data science and advanced analytics (DSAA). IEEE, 2016, pp. 130–139

    work page 2016

  3. [3]

    Aads: Augmented autonomous driving simulation using data-driven algorithms,

    W. Li, C. Pan, R. Zhang, J. Ren, Y . Ma, J. Fang, F. Yan, Q. Geng, X. Huang, H. Gonget al., “Aads: Augmented autonomous driving simulation using data-driven algorithms,”Science robotics, vol. 4, no. 28, p. eaaw0863, 2019

    work page 2019

  4. [4]

    Erasure-based interaction network for red-green-blue and thermal object detection and a unified benchmark,

    Q. Wang, Z. Tu, C. Li, H. Wang, and K. Wang, “Erasure-based interaction network for red-green-blue and thermal object detection and a unified benchmark,”Engineering Applications of Artificial Intelligence, vol. 157, p. 111155, 2025. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 10

    work page 2025

  5. [5]

    High performance rgb-thermal video object detection via hybrid fusion with progressive interaction and temporal-modal difference,

    Q. Wang, Z. Tu, C. Li, and J. Tang, “High performance rgb-thermal video object detection via hybrid fusion with progressive interaction and temporal-modal difference,”Information Fusion, vol. 114, p. 102665, 2025

    work page 2025

  6. [6]

    Performance evaluation of texture measures with classification based on kullback discrimination of distributions,

    T. Ojala, M. Pietikainen, and D. Harwood, “Performance evaluation of texture measures with classification based on kullback discrimination of distributions,” inProceedings of 12th international conference on pattern recognition, vol. 1. IEEE, 1994, pp. 582–585

    work page 1994

  7. [7]

    Cofnet: contrastive object-aware fusion using box-level masks for multispectral object detection,

    M. Zhou, Y . Li, G. Yang, X. Wei, H. Pu, J. Luo, and W. Jia, “Cofnet: contrastive object-aware fusion using box-level masks for multispectral object detection,”IEEE Transactions on Multimedia, 2025

    work page 2025

  8. [8]

    Amfd: Distillation via adaptive multimodal fusion for multispectral pedestrian detection,

    Z. Chen, Y . Qian, X. Yang, C. Wang, and M. Yang, “Amfd: Distillation via adaptive multimodal fusion for multispectral pedestrian detection,” IEEE Transactions on Multimedia, 2025

    work page 2025

  9. [9]

    Ssmpd: Semi-supervised learning for multispectral pedestrian detection,

    S. Shin, C. Lee, G.-M. Park, and J. U. Kim, “Ssmpd: Semi-supervised learning for multispectral pedestrian detection,”IEEE Transactions on Multimedia, 2025

    work page 2025

  10. [10]

    Dwsf-net: A dynamic wavelet-based spatial-frequency fusion network for multispectral object detection,

    F. Yang, W. Li, L. Li, M. Yang, and J. Zhang, “Dwsf-net: A dynamic wavelet-based spatial-frequency fusion network for multispectral object detection,”IEEE Transactions on Multimedia, 2026

    work page 2026

  11. [11]

    Cross-Modality Fusion Transformer for Multispectral Object Detection

    F. Qingyun, H. Dapeng, and W. Zhaokui, “Cross-modality fu- sion transformer for multispectral object detection,”arXiv preprint arXiv:2111.00273, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  12. [12]

    Icafusion: Iterative cross-attention guided feature fusion for multispectral object detection,

    J. Shen, Y . Chen, Y . Liu, X. Zuo, H. Fan, and W. Yang, “Icafusion: Iterative cross-attention guided feature fusion for multispectral object detection,”Pattern Recognition, vol. 145, p. 109913, 2024

    work page 2024

  13. [13]

    Removal then selection: A coarse-to-fine fusion perspective for rgb-infrared object detection,

    T. Zhao, M. Yuan, F. Jiang, N. Wang, and X. Wei, “Removal then selection: A coarse-to-fine fusion perspective for rgb-infrared object detection,”IEEE Transactions on Intelligent Transportation Systems, 2025

    work page 2025

  14. [14]

    Ei 2 det: Edge-guided illumination-aware interactive learning for visible-infrared object detection,

    K. Hu, Y . He, Y . Li, J. Zhao, S. Chen, and Y . Kang, “Ei 2 det: Edge-guided illumination-aware interactive learning for visible-infrared object detection,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

    work page 2025

  15. [15]

    Fusion-mamba for cross-modality object detection,

    W. Dong, H. Zhu, S. Lin, X. Luo, Y . Shen, G. Guo, and B. Zhang, “Fusion-mamba for cross-modality object detection,”IEEE Transactions on Multimedia, 2025

    work page 2025

  16. [16]

    Vmamba: Visual state space model,

    Y . Liu, Y . Tian, Y . Zhao, H. Yu, L. Xie, Y . Wang, Q. Ye, J. Jiao, and Y . Liu, “Vmamba: Visual state space model,”Advances in neural information processing systems, vol. 37, pp. 103 031–103 063, 2024

    work page 2024

  17. [17]

    Reflectance-guided progressive feature alignment network for all-day uav object detection,

    Z. Zhao, W. Zhang, Y . Xiao, C. Li, and J. Tang, “Reflectance-guided progressive feature alignment network for all-day uav object detection,” IEEE Transactions on Geoscience and Remote Sensing, 2025

    work page 2025

  18. [18]

    Deep feature flow for video recognition,

    X. Zhu, Y . Xiong, J. Dai, L. Yuan, and Y . Wei, “Deep feature flow for video recognition,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 2349–2358

    work page 2017

  19. [19]

    Flow-guided feature aggregation for video object detection,

    X. Zhu, Y . Wang, J. Dai, L. Yuan, and Y . Wei, “Flow-guided feature aggregation for video object detection,” inProceedings of the IEEE International Conference on Computer Vision, 2017, pp. 408–417

    work page 2017

  20. [20]

    Memory enhanced global- local aggregation for video object detection,

    Y . Chen, Y . Cao, H. Hu, and L. Wang, “Memory enhanced global- local aggregation for video object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 10 337–10 346

    work page 2020

  21. [21]

    Ultralytics YOLO,

    G. Jocher, J. Qiu, and A. Chaurasia, “Ultralytics YOLO,” Jan. 2023. [Online]. Available: https://github.com/ultralytics/ultralytics

    work page 2023

  22. [22]

    Hyper-yolo: When visual object detection meets hypergraph computation,

    Y . Feng, J. Huang, S. Du, S. Ying, J.-H. Yong, Y . Li, G. Ding, R. Ji, and Y . Gao, “Hyper-yolo: When visual object detection meets hypergraph computation,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    work page 2024

  23. [23]

    Mavic 3,

    DJI, “Mavic 3,” https://enterprise.dji.com/cn/mavic-3-enterprise/specs, 2025, accessed: 2025-06-05

    work page 2025

  24. [24]

    The pascal visual object classes (voc) challenge,

    M. Everingham, L. Van Gool, C. K. Williams, J. Winn, and A. Zisser- man, “The pascal visual object classes (voc) challenge,”International journal of computer vision, vol. 88, no. 2, pp. 303–338, 2010

    work page 2010

  25. [25]

    Microsoft coco: Common objects in context,

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” inComputer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13. Springer, 2014, pp. 740–755

    work page 2014

  26. [26]

    Kaist multi-spectral day/night data set for autonomous and as- sisted driving,

    Y . Choi, N. Kim, S. Hwang, K. Park, J. S. Yoon, K. An, and I. S. Kweon, “Kaist multi-spectral day/night data set for autonomous and as- sisted driving,”IEEE Transactions on Intelligent Transportation Systems, vol. 19, no. 3, pp. 934–948, 2018

    work page 2018

  27. [27]

    Drone-based rgb-infrared cross- modality vehicle detection via uncertainty-aware learning,

    Y . Sun, B. Cao, P. Zhu, and Q. Hu, “Drone-based rgb-infrared cross- modality vehicle detection via uncertainty-aware learning,”IEEE Trans- actions on Circuits and Systems for Video Technology, vol. 32, no. 10, pp. 6700–6713, 2022

    work page 2022

  28. [28]

    Vehicle detection in aerial imagery: A small target detection benchmark,

    S. Razakarivony and F. Jurie, “Vehicle detection in aerial imagery: A small target detection benchmark,”Journal of Visual Communication and Image Representation, vol. 34, pp. 187–203, 2016

    work page 2016

  29. [29]

    [Online]

    Cvc-14: Visible-fir day-night pedestrian sequence dataset. [Online]. Available: https://adas.cvc.uab.es/elektra/enigma-portfolio/ cvc-14-visible-fir-day-night-pedestrian-sequence-dataset/

    work page

  30. [30]

    Pedestrian detection at day/night time with visible and fir cameras: A comparison,

    A. Gonz ´alez, Z. Fang, Y . Socarras, J. Serrat, D. V ´azquez, J. Xu, and A. M. L ´opez, “Pedestrian detection at day/night time with visible and fir cameras: A comparison,”Sensors, vol. 16, no. 6, p. 820, 2016

    work page 2016

  31. [31]

    Llvip: A visible-infrared paired dataset for low-light vision,

    X. Jia, C. Zhu, M. Li, W. Tang, and W. Zhou, “Llvip: A visible-infrared paired dataset for low-light vision,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 3496–3504

    work page 2021

  32. [32]

    Target-aware dual adversarial learning and a multi-scenario multi- modality benchmark to fuse infrared and visible for object detection,

    J. Liu, X. Fan, Z. Huang, G. Wu, R. Liu, W. Zhong, and Z. Luo, “Target-aware dual adversarial learning and a multi-scenario multi- modality benchmark to fuse infrared and visible for object detection,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5802–5811

    work page 2022

  33. [33]

    Misaligned visible-thermal object detection: A drone-based benchmark and base- line,

    K. Song, X. Xue, H. Wen, Y . Ji, Y . Yan, and Q. Meng, “Misaligned visible-thermal object detection: A drone-based benchmark and base- line,”IEEE Transactions on Intelligent Vehicles, 2024

    work page 2024

  34. [34]

    Faster r-cnn: Towards real-time object detection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: Towards real-time object detection with region proposal networks,”IEEE transactions on pattern analysis and machine intelligence, vol. 39, no. 6, pp. 1137–1149, 2016

    work page 2016

  35. [35]

    YOLOv10: Real-Time End-to-End Object Detection

    A. Wang, H. Chen, L. Liu, K. Chen, Z. Lin, J. Han, and G. Ding, “Yolov10: Real-time end-to-end object detection,”arXiv preprint arXiv:2405.14458, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  36. [36]

    Tood: Task- aligned one-stage object detection,

    C. Feng, Y . Zhong, Y . Gao, M. R. Scott, and W. Huang, “Tood: Task- aligned one-stage object detection,” inICCV, 2021

    work page 2021

  37. [37]

    Deformable detr: Deformable transformers for end-to-end object detection,

    X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable detr: Deformable transformers for end-to-end object detection,” in International Conference on Learning Representations, 2021. [Online]. Available: https://openreview.net/forum?id=gZ9hCDWe6ke

    work page 2021

  38. [38]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 2980–2988

    work page 2017

  39. [39]

    Dino: Detr with improved denoising anchor boxes for end-to- end object detection,

    H. Zhang, F. Li, S. Liu, L. Zhang, H. Su, J. Zhu, L. M. Ni, and H.-Y . Shum, “Dino: Detr with improved denoising anchor boxes for end-to- end object detection,” 2022

    work page 2022

  40. [40]

    Dense distinct query for end-to-end object detection,

    S. Zhang, X. Wang, J. Wang, J. Pang, C. Lyu, W. Zhang, P. Luo, and K. Chen, “Dense distinct query for end-to-end object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 7329–7338

    work page 2023

  41. [41]

    M-specgene: Generalized foundation model for rgbt multispec- tral vision,

    K. Zhou, F. Yang, S. Wang, B. Wen, C. Zi, L. Chen, Q. Shen, and X. Cao, “M-specgene: Generalized foundation model for rgbt multispec- tral vision,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 7861–7872

    work page 2025

  42. [42]

    Relation distillation networks for video object detection,

    J. Deng, Y . Pan, T. Yao, W. Zhou, H. Li, and T. Mei, “Relation distillation networks for video object detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 7023–7032

    work page 2019

  43. [43]

    YOLOv3: An Incremental Improvement

    J. Redmon and A. Farhadi, “Yolov3: An incremental improvement,” arXiv preprint arXiv:1804.02767, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  44. [44]

    YOLOv9: Learning What You Want to Learn Using Programmable Gradient Information

    C.-Y . Wang, I.-H. Yeh, and H.-Y . M. Liao, “Yolov9: Learning what you want to learn using programmable gradient information,”arXiv preprint arXiv:2402.13616, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  45. [45]

    Detrs beat yolos on real-time object detection,

    Y . Zhao, W. Lv, S. Xu, J. Wei, G. Wang, Q. Dang, Y . Liu, and J. Chen, “Detrs beat yolos on real-time object detection,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 16 965–16 974

    work page 2024

  46. [46]

    Diffusiondet: Diffusion model for object detection,

    S. Chen, P. Sun, Y . Song, and P. Luo, “Diffusiondet: Diffusion model for object detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 19 830–19 843

    work page 2023

  47. [47]

    Sequence level semantics aggregation for video object detection,

    H. Wu, Y . Chen, N. Wang, and Z. Zhang, “Sequence level semantics aggregation for video object detection,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 9217–9225

    work page 2019

  48. [48]

    Temporal roi align for video object recognition,

    T. Gong, K. Chen, X. Wang, Q. Chu, F. Zhu, D. Lin, N. Yu, and H. Feng, “Temporal roi align for video object recognition,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 2, 2021, pp. 1442–1450

    work page 2021

  49. [49]

    A new dataset and a baseline model for breast lesion detection in ultrasound videos,

    Z. Lin, J. Lin, L. Zhu, H. Fu, J. Qin, and L. Wang, “A new dataset and a baseline model for breast lesion detection in ultrasound videos,” inMedical Image Computing and Computer Assisted Intervention – MICCAI 2022, L. Wang, Q. Dou, P. T. Fletcher, S. Speidel, and S. Li, Eds. Cham: Springer Nature Switzerland, 2022, pp. 614–623

    work page 2022

  50. [50]

    A spatial-temporal deformable attention based framework for breast lesion detection in JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 11 videos,

    C. Qin, J. Cao, H. Fu, R. M. Anwer, and F. S. Khan, “A spatial-temporal deformable attention based framework for breast lesion detection in JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 11 videos,” inInternational Conference on Medical Image Computing and Computer-Assisted Intervention. Springer, 2023, pp. 479–488

    work page 2021

  51. [51]

    Mixture of Scale Experts for Alignment-free RGBT Video Object Detection and A Unified Benchmark

    Q. Wang, Z. Tu, K. Wang, L. Gu, and C. Guo, “Mixture of scale experts for alignment-free rgbt video object detection and a unified benchmark,” 2025. [Online]. Available: https://arxiv.org/abs/2410.12143

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  52. [52]

    Cat: Cross attention in vision transformer,

    H. Lin, X. Cheng, X. Wu, and D. Shen, “Cat: Cross attention in vision transformer,” in2022 IEEE international conference on multimedia and expo (ICME). IEEE, 2022, pp. 1–6

    work page 2022

  53. [53]

    Dcevo: Discriminative cross-dimensional evolutionary learning for infrared and visible image fusion,

    J. Liu, B. Zhang, Q. Mei, X. Li, Y . Zou, Z. Jiang, L. Ma, R. Liu, and X. Fan, “Dcevo: Discriminative cross-dimensional evolutionary learning for infrared and visible image fusion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025, pp. 2226–2235

    work page 2025

  54. [54]

    Task-driven image fusion with learnable fusion loss,

    H. Bai, J. Zhang, Z. Zhao, Y . Wu, L. Deng, Y . Cui, T. Feng, and S. Xu, “Task-driven image fusion with learnable fusion loss,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 7457–7468

    work page 2025