pith. machine review for the scientific record. sign in

arxiv: 2605.06084 · v1 · submitted 2026-05-07 · 💻 cs.CV

Recognition: unknown

AMIEOD: Adaptive Multi-Experts Image Enhancement for Object Detection in Low-Illumination Scenes

Xiaochen Huang , Honggang Chen , Weicheng Zhang , Xiaobo Dai , Yongyi Li , Linbo Qing , Xiaohai He
Authors on Pith no claims yet

Pith reviewed 2026-05-08 14:15 UTC · model grok-4.3

classification 💻 cs.CV
keywords low-illuminationimage enhancementobject detectionmulti-expert moduledetection-guided lossadaptive selectionjoint optimization
0
0 comments X

The pith

A multi-expert enhancement system trained with detection-guided losses improves object detection accuracy on low-illumination images.

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

The paper introduces a framework that treats image enhancement and object detection as a single joint task rather than separate steps. A module offers several enhancement strategies at once, while losses derived from the detector's own outputs steer which strategy is chosen and how the enhancement is regressed. This alignment is meant to ensure that the enhanced image actually helps the downstream detector instead of optimizing for visual appeal alone. The approach is designed to slot into existing detectors without changing their architecture. Experiments across datasets show higher detection scores in dim conditions than either raw low-light inputs or conventional enhancement pipelines.

Core claim

The central claim is that a Multi-Experts Image Enhancement Module, optimized through a Detection-Guided Regression Loss that sets regression targets from detection results and a Detection-Guided Cross-Entropy loss that turns expert selection into a classification problem, produces enhancement choices that raise detection performance when paired with standard detectors on low-illumination scenes.

What carries the argument

The Multi-Experts Image Enhancement Module (MEIEM) together with the Expert Selection Module (ESM), steered by Detection-Guided Regression Loss (DGRL) and Detection-Guided Cross-Entropy (DGCE) loss, which make enhancement decisions depend directly on how well the detector performs afterward.

If this is right

  • The framework can be attached to existing object detectors to raise their accuracy in dim scenes without architectural changes.
  • During inference the Expert Selection Module picks the most suitable enhancement expert on a per-image basis using the learned classification signal.
  • Information in poorly lit images is exploited more effectively because enhancement targets are set by detection outcomes rather than generic image quality metrics.
  • The joint optimization produces enhancement that is already tuned to the needs of the downstream detection task.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same detection-guided selection idea could be tested on related low-quality inputs such as motion-blurred or noisy images to see whether the alignment benefit transfers.
  • If expert selection proves reliable across illumination levels, the method might reduce reliance on separately trained low-light detectors.
  • The approach implicitly assumes that multiple enhancement experts provide complementary information; removing the multi-expert structure would test whether a single learned enhancer suffices.

Load-bearing premise

The premise that losses guided by detector outputs will generate enhancement strategies that genuinely raise detection scores rather than create new artifacts or biases the detector can exploit during training.

What would settle it

If the full AMIEOD pipeline, after joint training, is evaluated on standard low-illumination detection benchmarks and yields no higher mean average precision than the same detector run on the original unenhanced images, the claimed improvement would not hold.

Figures

Figures reproduced from arXiv: 2605.06084 by Honggang Chen, Linbo Qing, Weicheng Zhang, Xiaobo Dai, Xiaochen Huang, Xiaohai He, Yongyi Li.

Figure 1
Figure 1. Figure 1: General structure of different low-illumination image view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our proposed method. “FC” is the full connection layer. The training process consists of two stages. The view at source ↗
Figure 4
Figure 4. Figure 4: The structure of IAEM. “PP” is the parameter view at source ↗
Figure 3
Figure 3. Figure 3: The model structure of PIEM and JIEM. In MEIEM, we incorporate a Pretrained Image Enhancement Module (PIEM), a Jointly Optimized Enhancement Module (JIEM), and an Image-Adaptive Enhancement Module (IAEM). PIEM adopts the pretrained SCI (Self-Calibrated Illumination) model [9], whose parameters are frozen to enhance image brightness and provide clearer object cues. JIEM shares the same architecture as PIEM,… view at source ↗
Figure 5
Figure 5. Figure 5: Visualization results on Exdark dataset. The areas outlined by red dashed lines are enlarged for better visualization. view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of heatmap visualization results. All comparison algorithms adopt YOLOv3 as base detector. view at source ↗
Figure 7
Figure 7. Figure 7: The performances on different baselines. view at source ↗
Figure 9
Figure 9. Figure 9: Single- and multi-enhancement training strategies. view at source ↗
Figure 10
Figure 10. Figure 10: Detection loss trends during MEIEM and detector training. Horizontal axis refers the number of epoch. Vertical axis view at source ↗
Figure 11
Figure 11. Figure 11: Variation of the DGCE loss during ESM training stage view at source ↗
Figure 12
Figure 12. Figure 12: Visualization results of false case. V. CONCLUSION In this study, we propose a detector-flexible object detection approach for low-illumination image, denoted by AMIEOD, It first introduces MEIEM to enhance the image from four distinct perspectives, thereby comprehensively exploiting image information for model training to improve detection performance in poorly lit images. To better align MEIEM with the … view at source ↗
read the original abstract

In multimedia application scenarios, images captured under low-illumination conditions often lead to lower accuracy in visual perception tasks compared to those taken in well-lit environments. To tackle this challenge, we propose AMIEOD, an image enhancement-enabled object detection framework for low-illumination scenes, where the two tasks are jointly optimized in a detection performance-oriented manner. Specifically, to fully exploit the information in poorly lit images, a Multi-Experts Image Enhancement Module (MEIEM) is proposed, which leverages diverse enhancement strategies. On this basis, aiming to better align the MEIEM with the detection task, we propose a Detection-Guided Regression Loss (DGRL) that utilizes the detection result to decide the regression target. Moreover, to dynamically select the most suitable enhancement strategy from MEIEM during inference, we construct an Expert Selection Module (ESM) guided by the proposed Detection-Guided Cross-Entropy (DGCE) loss, which formulates the optimization of ESM as a classification task. The improved method is well-matched with current detection algorithms to improve their performance in dim scenes. Extensive experiments on multiple datasets demonstrate that the proposed method significantly improves object detection accuracy in low-illumination conditions. Our code has been released at https://github.com/scujayfantasy/AMIEOD

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes AMIEOD, a joint optimization framework for low-illumination image enhancement and object detection. It introduces a Multi-Experts Image Enhancement Module (MEIEM) that applies diverse enhancement strategies, a Detection-Guided Regression Loss (DGRL) that uses detection outputs to set regression targets, and an Expert Selection Module (ESM) trained with Detection-Guided Cross-Entropy (DGCE) loss to choose the best expert at inference. The method is designed to be compatible with existing detectors and is evaluated on multiple datasets, with the central claim being that it significantly boosts detection accuracy in dim scenes. Code is released.

Significance. If the reported gains prove robust and not attributable to detector-specific shortcuts, the multi-expert enhancement with detection-guided supervision could offer a practical plug-in improvement for real-world low-light detection tasks such as surveillance or autonomous navigation. The explicit release of code supports reproducibility, which strengthens the contribution.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (method): The DGRL and DGCE losses are defined directly from the detector's regression and classification outputs. This creates a plausible incentive for MEIEM to generate images whose statistics match the detector's training distribution even if they contain unnatural artifacts. No frozen-detector ablation, cross-detector transfer experiment, or perceptual-quality metric on the enhanced images is described to rule out such exploitation, which is load-bearing for the claim that the method produces genuine visibility improvement rather than detector-specific cues.
  2. [Experiments] Experiments section: The abstract states that extensive experiments on multiple datasets demonstrate significant accuracy gains, yet the provided description contains no quantitative tables, ablation results on the individual losses or expert count, or error analysis showing where the method fails. Without these, it is impossible to verify whether the central performance claim holds or whether post-hoc design choices inflated the reported improvements.
minor comments (2)
  1. [Abstract] The abstract mentions that the method is 'well-matched with current detection algorithms,' but does not specify which detectors were tested or whether the joint training requires retraining the detector from scratch.
  2. Notation for the expert selection and loss weighting is introduced without an explicit equation reference or diagram in the method overview, making the flow from MEIEM to ESM harder to follow on first reading.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to strengthen the experimental validation and address concerns about potential detector-specific effects.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (method): The DGRL and DGCE losses are defined directly from the detector's regression and classification outputs. This creates a plausible incentive for MEIEM to generate images whose statistics match the detector's training distribution even if they contain unnatural artifacts. No frozen-detector ablation, cross-detector transfer experiment, or perceptual-quality metric on the enhanced images is described to rule out such exploitation, which is load-bearing for the claim that the method produces genuine visibility improvement rather than detector-specific cues.

    Authors: We acknowledge the validity of this concern. Although MEIEM applies standard enhancement operations, the detection-guided losses could in principle encourage detector-specific cues. In the revised manuscript we will add (1) a frozen-detector ablation in which the detector remains fixed while only the enhancement module is trained, (2) cross-detector transfer results applying the learned enhancement to YOLOv5 and Faster R-CNN, and (3) perceptual-quality scores (NIQE, BRISQUE) on the enhanced images. These additions will help demonstrate that performance gains arise from improved visibility rather than exploitation. revision: yes

  2. Referee: [Experiments] Experiments section: The abstract states that extensive experiments on multiple datasets demonstrate significant accuracy gains, yet the provided description contains no quantitative tables, ablation results on the individual losses or expert count, or error analysis showing where the method fails. Without these, it is impossible to verify whether the central performance claim holds or whether post-hoc design choices inflated the reported improvements.

    Authors: We apologize for the insufficient detail in the submitted version. We will expand the Experiments section with (i) full quantitative tables reporting mAP gains on ExDark, DarkFace and additional low-light datasets, (ii) ablation tables isolating the contribution of DGRL, DGCE and the number of experts in MEIEM, and (iii) an error-analysis subsection discussing failure cases under extreme low illumination. These revisions will make the performance claims fully verifiable. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper defines MEIEM, DGRL, and DGCE as components in a joint optimization where detection outputs supervise the enhancement module. This is a standard end-to-end training setup rather than any quantity being defined in terms of itself or a fitted parameter being renamed as a prediction. No equations or sections in the provided abstract reduce the claimed performance gains to the inputs by construction. The central claim rests on experimental results across datasets, which are external to the derivation and constitute independent evidence.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available, so no concrete free parameters, axioms, or invented physical entities can be extracted. The new algorithmic components (MEIEM, DGRL, ESM) are design choices rather than postulated entities with independent evidence.

pith-pipeline@v0.9.0 · 5561 in / 1146 out tokens · 67577 ms · 2026-05-08T14:15:33.528439+00:00 · methodology

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

60 extracted references · 3 canonical work pages · 2 internal anchors

  1. [1]

    A review of object detection: Datasets, performance evaluation, architecture, applications and current trends,

    W. Chen, J. Luo, F. Zhang, and Z. Tian, “A review of object detection: Datasets, performance evaluation, architecture, applications and current trends,”Multimedia Tools and Applications, vol. 83, no. 24, pp. 65 603–65 661, 2024

    2024

  2. [2]

    An adaptive sample assignment network for tiny object detection,

    H. Dai, S. Gao, H. Huang, D. Mao, C. Zhang, and Y . Zhou, “An adaptive sample assignment network for tiny object detection,”IEEE Transactions on Multimedia, vol. 26, pp. 2918–2931, 2023

    2023

  3. [3]

    Frfcnet: Feature refinement and flexible concatenation for object detection,

    T. Zhang, Z. Wu, X. He, and Q. Wu, “Frfcnet: Feature refinement and flexible concatenation for object detection,”IEEE Transactions on Multimedia, pp. 1–12, 2025

    2025

  4. [4]

    Degradation modeling for restoration-enhanced object detection in adverse weather scenes,

    X. Wang, X. Liu, H. Yang, Z. Wang, X. Wen, X. He, L. Qing, and H. Chen, “Degradation modeling for restoration-enhanced object detection in adverse weather scenes,”IEEE Transactions on Intelligent Vehicles, 2024

    2024

  5. [5]

    Adaptive and background-aware vision transformer for real-time uav tracking,

    S. Li, Y . Yang, D. Zeng, and X. Wang, “Adaptive and background-aware vision transformer for real-time uav tracking,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 13 989–14 000

    2023

  6. [6]

    A lightweight framework for robust object detection in adverse weather based on dual-teacher feature alignment,

    R. Hu, H. Zheng, S. Ye, L. Qing, and H. Chen, “A lightweight framework for robust object detection in adverse weather based on dual-teacher feature alignment,”Neurocomputing, vol. 671, p. 132726, 2026

    2026

  7. [7]

    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,” inEuropean Conference on Computer Vision. Springer, 2014, pp. 740–755

    2014

  8. [8]

    Pascal voc 2008 challenge,

    D. Hoiem, S. K. Divvala, and J. H. Hays, “Pascal voc 2008 challenge,” World Literature Today, vol. 24, no. 1, pp. 1–4, 2009

    2008

  9. [9]

    Toward fast, flexible, and robust low-light image enhancement,

    L. Ma, T. Ma, R. Liu, X. Fan, and Z. Luo, “Toward fast, flexible, and robust low-light image enhancement,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 5637–5646

    2022

  10. [10]

    Rethinking image resation for object detection,

    S. Sun, W. Ren, T. Wang, and X. Cao, “Rethinking image resation for object detection,”Advances in Neural Information Processing Systems, vol. 35, pp. 4461–4474, 2022

    2022

  11. [11]

    Pe-yolo: Pyramid enhancement network for dark object detection,

    X. Yin, Z. Yu, Z. Fei, W. Lv, and X. Gao, “Pe-yolo: Pyramid enhancement network for dark object detection,” inInternational Conference on Artificial Neural Networks. Springer, 2023, pp. 163–174

    2023

  12. [12]

    Denet: Detection-driven enhancement network for object detection under adverse weather conditions,

    Q. Qin, K. Chang, M. Huang, and G. Li, “Denet: Detection-driven enhancement network for object detection under adverse weather conditions,” inProceedings of the Asian Conference on Computer Vision, 2022, pp. 2813–2829

    2022

  13. [13]

    Image-adaptive yolo for object detection in adverse weather conditions,

    W. Liu, G. Ren, R. Yu, S. Guo, J. Zhu, and L. Zhang, “Image-adaptive yolo for object detection in adverse weather conditions,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 36, no. 2, 2022, pp. 1792–1800

    2022

  14. [14]

    Gdip: Gated differentiable image processing for object detection in adverse conditions,

    S. Kalwar, D. Patel, A. Aanegola, K. R. Konda, S. Garg, and K. M. Krishna, “Gdip: Gated differentiable image processing for object detection in adverse conditions,” in2023 IEEE International Conference on Robotics and Automation. IEEE, 2023, pp. 7083–7089

    2023

  15. [15]

    Erup-yolo: Enhancing object detection robustness for adverse weather condition by unified image-adaptive processing,

    Y . Ogino, Y . Shoji, T. Toizumi, and A. Ito, “Erup-yolo: Enhancing object detection robustness for adverse weather condition by unified image-adaptive processing,” in2025 IEEE/CVF Winter Conference on Applications of Computer Vision. IEEE, 2025, pp. 8597–8605

    2025

  16. [16]

    Toward highly efficient semantic-guided machine vision for low-light object detection,

    X. Feng, J. Zeng, S. Wang, and Z. He, “Toward highly efficient semantic-guided machine vision for low-light object detection,” in35th British Machine Vision Conference, 2024, pp. 25–28. MANUSCRIPT SUBMITTED TO IEEE TRANSACTIONS ON MULTIMEDIA, OCT., 2025 14

    2024

  17. [17]

    Lightstar-net: A pseudo-raw space enhancement for efficient low-light object detection,

    X. Feng, J. Wang, S. Wang, and J. Zhang, “Lightstar-net: A pseudo-raw space enhancement for efficient low-light object detection,” inInternational Conference on Computational Visual Media. Springer, 2025, pp. 192–211

    2025

  18. [18]

    Fcma-det: Low-light image object detection based on feature complementarity and multi-content aggregation,

    J. Ji, Y . Zhao, Y . Zhang, X. Zuo, C. Wang, and F. Shi, “Fcma-det: Low-light image object detection based on feature complementarity and multi-content aggregation,”IEEE Transactions on Geoscience and Remote Sensing, 2025

    2025

  19. [19]

    Trash to treasure: Low-light object detection via decomposition-and-aggregation,

    X. Cui, L. Ma, T. Ma, J. Liu, X. Fan, and R. Liu, “Trash to treasure: Low-light object detection via decomposition-and-aggregation,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 2, 2024, pp. 1417–1425

    2024

  20. [20]

    Multitask aet with orthogonal tangent regularity for dark object detection,

    Z. Cui, G.-J. Qi, L. Gu, S. You, Z. Zhang, and T. Harada, “Multitask aet with orthogonal tangent regularity for dark object detection,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021, pp. 2553–2562

    2021

  21. [21]

    2pcnet: Two-phase consistency training for day-to-night unsupervised domain adaptive object detection,

    M. Kennerley, J.-G. Wang, B. Veeravalli, and R. T. Tan, “2pcnet: Two-phase consistency training for day-to-night unsupervised domain adaptive object detection,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 11 484–11 493

    2023

  22. [22]

    Isp-teacher: Image signal process with disentanglement regularization for unsupervised domain adaptive dark object detection,

    Y . Zhang, Y . Zhang, Z. Zhang, M. Zhang, R. Tian, and M. Ding, “Isp-teacher: Image signal process with disentanglement regularization for unsupervised domain adaptive dark object detection,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 7, 2024, pp. 7387–7395

    2024

  23. [23]

    Boosting object detection with zero-shot day-night domain adaptation,

    Z. Du, M. Shi, and J. Deng, “Boosting object detection with zero-shot day-night domain adaptation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 12 666–12 676

    2024

  24. [24]

    You only look around: Learning illumination-invariant feature for low-light object detection,

    M. Hong, S. Cheng, H. Huang, H. Fan, and S. Liu, “You only look around: Learning illumination-invariant feature for low-light object detection,”Advances in Neural Information Processing Systems, vol. 37, pp. 87 136–87 158, 2024

    2024

  25. [25]

    Mbllen: Low-light image/video enhancement using cnns

    F. Lv, F. Lu, J. Wu, and C. Lim, “Mbllen: Low-light image/video enhancement using cnns.” inBritish Machine Vision Conference, vol. 220, no. 1. Northumbria University, 2018, p. 4

    2018

  26. [26]

    Kindling the darkness: A practical low-light image enhancer,

    Y . Zhang, J. Zhang, and X. Guo, “Kindling the darkness: A practical low-light image enhancer,” inProceedings of the 27th ACM International Conference on Multimedia, 2019, pp. 1632–1640

    2019

  27. [27]

    Zero-reference deep curve estimation for low-light image enhancement,

    C. Guo, C. Li, J. Guo, C. C. Loy, J. Hou, S. Kwong, and R. Cong, “Zero-reference deep curve estimation for low-light image enhancement,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 1780–1789

    2020

  28. [28]

    Information maximizing adaptation network with label distribution priors for unsupervised domain adaptation,

    P. Wang, Y . Yang, Y . Xia, K. Wang, X. Zhang, and S. Wang, “Information maximizing adaptation network with label distribution priors for unsupervised domain adaptation,”IEEE Transactions on Multimedia, vol. 25, pp. 6026–6039, 2023

    2023

  29. [29]

    Dsll-face: Distributed supervision-integrated framework for low-light face detection,

    S. Chen, K. Chen, G. Wang, S. Wen, and Z. Zhou, “Dsll-face: Distributed supervision-integrated framework for low-light face detection,”IEEE Transactions on Multimedia, pp. 1–12, 2025

    2025

  30. [30]

    Recurrent exposure generation for low-light face detection,

    J. Liang, J. Wang, Y . Quan, T. Chen, J. Liu, H. Ling, and Y . Xu, “Recurrent exposure generation for low-light face detection,”IEEE Transactions on Multimedia, vol. 24, pp. 1609–1621, 2022

    2022

  31. [31]

    Ubtransformer: Uncertainty-based transformer model for complex scenarios detection in autonomous driving,

    K. Wang, Q. Ma, X. Li, C. Shen, R. Leng, and J. Lu, “Ubtransformer: Uncertainty-based transformer model for complex scenarios detection in autonomous driving,”IEEE Transactions on Multimedia, pp. 1–11, 2025

    2025

  32. [32]

    Yolosr-ist: A deep learning method for small target detection in infrared remote sensing images based on super-resolution and yolo,

    R. Li and Y . Shen, “Yolosr-ist: A deep learning method for small target detection in infrared remote sensing images based on super-resolution and yolo,”Signal Processing, vol. 208, p. 108962, 2023

    2023

  33. [33]

    Rich feature hierarchies for accurate object detection and semantic segmentation,

    R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 580–587

    2014

  34. [34]

    Fast r-cnn,

    R. Girshick, “Fast r-cnn,” inProceedings of the IEEE international conference on computer vision, 2015, pp. 1440–1448

    2015

  35. [35]

    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,”Advances in neural information processing systems, vol. 28, 2015

    2015

  36. [36]

    Mask r-cnn,

    K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2961–2969

    2017

  37. [37]

    Superyolo: Super resolution assisted object detection in multimodal remote sensing imagery,

    J. Zhang, J. Lei, W. Xie, Z. Fang, Y . Li, and Q. Du, “Superyolo: Super resolution assisted object detection in multimodal remote sensing imagery,”IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–15, 2023

    2023

  38. [38]

    Crkd-yolo: Cross-resolution knowledge distillation for low-resolution remote sensing image object detection,

    X. Huang, Q. Teng, H. Yang, X. He, L. Qing, P. Wang, and H. Chen, “Crkd-yolo: Cross-resolution knowledge distillation for low-resolution remote sensing image object detection,”IEEE Transactions on Instrumentation and Measurement, 2025

    2025

  39. [39]

    Frfcnet: Feature refinement and flexible concatenation for object detection,

    T. Zhang, Z. Wu, X. He, and Q. Wu, “Frfcnet: Feature refinement and flexible concatenation for object detection,”IEEE Transactions on Multimedia, 2025

    2025

  40. [40]

    Ssd: Single shot multibox detector,

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y . Fu, and A. C. Berg, “Ssd: Single shot multibox detector,” inEuropean conference on computer vision. Springer, 2016, pp. 21–37

    2016

  41. [41]

    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

    2017

  42. [42]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 2117–2125

    2017

  43. [43]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in neural information processing systems, vol. 30, 2017

    2017

  44. [44]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in European conference on computer vision. Springer, 2020, pp. 213–229

    2020

  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

    2024

  46. [46]

    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,”arXiv preprint arXiv:2203.03605, 2022

    work page internal anchor Pith review arXiv 2022

  47. [47]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning. PmLR, 2021, pp. 8748–8763

    2021

  48. [48]

    A survey of zero-shot object detection,

    W. Cao, X. Yao, Z. Xu, Y . Liu, Y . Pan, and Z. Ming, “A survey of zero-shot object detection,”Big Data Mining and Analytics, vol. 8, no. 3, pp. 726–750, 2025

    2025

  49. [49]

    Ha-fgovd: Highlighting fine-grained attributes via explicit linear composition for open-vocabulary object detection,

    Y . Ma, M. Liu, C. Zhu, and X.-C. Yin, “Ha-fgovd: Highlighting fine-grained attributes via explicit linear composition for open-vocabulary object detection,”IEEE Transactions on Multimedia, vol. 27, pp. 3171–3183, 2025

    2025

  50. [50]

    Research on marker recognition method for substation engineering progress monitoring based on grounding dino,

    L. Ma, M. Zhou, Q. Wu, T. Zhang, H. Zhang, and J. Cai, “Research on marker recognition method for substation engineering progress monitoring based on grounding dino,” in2024 The 9th International Conference on Power and Renewable Energy (ICPRE), 2024, pp. 776–780

    2024

  51. [51]

    M2fnet: Mask-guided multi-level fusion for rgb-t pedestrian detection,

    X. Li, S. Chen, C. Tian, H. Zhou, and Z. Zhang, “M2fnet: Mask-guided multi-level fusion for rgb-t pedestrian detection,”IEEE Transactions on Multimedia, vol. 26, pp. 8678–8690, 2024

    2024

  52. [52]

    Detection-friendly dehazing: Object detection in real-world hazy scenes,

    C. Li, H. Zhou, Y . Liu, C. Yang, Y . Xie, Z. Li, and L. Zhu, “Detection-friendly dehazing: Object detection in real-world hazy scenes,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 7, pp. 8284–8295, 2023

    2023

  53. [53]

    Lightness and retinex theory,

    E. H. Land and J. J. McCann, “Lightness and retinex theory,”Journal of the Optical Society of America, vol. 61, no. 1, pp. 1–11, 1971

    1971

  54. [54]

    Getting to know low-light images with the exclusively dark dataset,

    Y . P. Loh and C. S. Chan, “Getting to know low-light images with the exclusively dark dataset,”Computer Vision and Image Understanding, vol. 178, pp. 30–42, 2019

    2019

  55. [55]

    Advancing image understanding in poor visibility environments: A collective benchmark study,

    W. Yang, Y . Yuan, W. Ren, J. Liu, W. J. Scheirer, Z. Wang, T. Zhang, Q. Zhong, D. Xie, S. Puet al., “Advancing image understanding in poor visibility environments: A collective benchmark study,”IEEE Transactions on Image Processing, vol. 29, pp. 5737–5752, 2020

    2020

  56. [56]

    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

    2021

  57. [57]

    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

    2022

  58. [58]

    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 arXiv 2018

  59. [59]

    Stochastic gradient descent tricks,

    L. Bottou, “Stochastic gradient descent tricks,” inNeural networks: tricks of the trade (2nd en.). Springer, 2012, pp. 421–436

    2012

  60. [60]

    Use hirescam instead of grad-cam for faithful explanations of convolutional neural networks,

    R. L. Draelos and L. Carin, “Use hirescam instead of grad-cam for faithful explanations of convolutional neural networks,”arXiv preprint arXiv:2011.08891, 2020

    work page arXiv 2011