pith. sign in

arxiv: 2606.30215 · v1 · pith:CBTKRO5Tnew · submitted 2026-06-29 · 💻 cs.CV · cs.AI

Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion

Chao Tian , Zikun Zhou , Chao Yang , Guoqing Zhu , Zhenyu He This is my paper

Pith reviewed 2026-06-30 06:14 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords RGB-T detectionsparse fusionefficient object detectioncross-modality fusiontwo-stage detectionregion proposalsthermal imaging
0
0 comments X

The pith

A sparse cross-modality fusion approach enables efficient RGB-T object detection by first identifying high-recall proposals with lightweight single-modality models and then fusing features only on those sparse regions.

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

RGB-T detectors combine visible and thermal images for robust object detection but often use expensive full-image fusion. The paper shows that most regions are simple backgrounds best handled by cheap single-modality checks. It introduces a two-stage process that first generates many candidate boxes with lightweight per-modality detectors and then fuses features only on those few candidates to refine results. This focuses heavy computation on likely objects, cutting overall cost while preserving accuracy and scaling to large images.

Core claim

The paper establishes that efficient RGB-T object detection is achievable through a sparse cross-modality fusion mechanism implemented as a two-stage framework. In the first stage, modality-specific lightweight detectors rapidly produce high-recall regions of interest across the image. The second stage then performs feature fusion and refinement exclusively on these sparse proposals, filtering false positives and adjusting bounding boxes. This allows the system to allocate computational resources adaptively to potential foreground areas rather than processing the entire image uniformly.

What carries the argument

The sparse fusion mechanism that restricts cross-modality feature fusion to high-recall RoIs identified by lightweight single-modality detectors.

If this is right

  • Competitive detection performance is maintained despite using substantially fewer parameters.
  • Computational cost is significantly reduced compared to exhaustive fusion methods.
  • The approach scales effectively to high-resolution input images.
  • Detection accuracy remains high under challenging conditions by leveraging complementary modality strengths only where needed.

Where Pith is reading between the lines

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

  • Similar sparse strategies could apply to other sensor fusion tasks in computer vision where background regions predominate.
  • The method might enable real-time RGB-T detection on resource-constrained devices like drones or mobile cameras.
  • If the first-stage recall drops in certain environments, overall system performance could degrade more than in dense-fusion baselines.

Load-bearing premise

The assumption that most image regions consist of smooth backgrounds easily processed by lightweight single-modality models without missing objects.

What would settle it

A test on images where backgrounds contain many false-object-like textures or small objects are distributed evenly would show if the lightweight first stage fails to achieve high recall, causing the overall detector to underperform full-fusion alternatives.

Figures

Figures reproduced from arXiv: 2606.30215 by Chao Tian, Chao Yang, Guoqing Zhu, Zhenyu He, Zikun Zhou.

Figure 1
Figure 1. Figure 1: Illustration for our motivation and insight. (a) Most of an image is background that can be efficiently filtered out without fusion. Sparse fusion for potential areas could avoid the computationally intensive dense fusion over the entire image. (b) The P-R curves of YOLOv8-Large and -Small suggest that, though the lightweight detector has more false positives and lower location accuracy, it maintains a hig… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the SFEDet framework that adopts the sparse fusion mecha￾nism. It consists of two main components: the dual lightweight RPNs and the elabo￾rate Fusion-driven Examination and Refinement (FER) module. The RPNs filter the background and identify potential foregrounds, i.e., the RoIs, across the entire im￾age. The FER module performs step-wise cross-modality fusion, classification, and box refi… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison with COFNet in challenging illumination conditions, demonstrating the effectiveness of our SFEDet. β = 7.5, and γ = 0.375. We adopt the AdamW optimizer for training, with a learning rate of 0.0002 and a batch size of 6. The learning rate decay for the RPNs, i.e., the λ, is set to 0.005. Please refer to Appendix B for more details. 4.2 Comparisons with State-of-the-art Methods Results… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison between FER and vanilla fusion. Effects of model components. The ablation study of different modules is sum￾marized in [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The relationship between precision/complexity and sparsity. The results on different benchmarks demonstrate that our detector maintains high precision even un￾der large sparsity, while significantly reducing computational cost, especially for high￾resolution images. Blue lines indicate the setting of our reported performance. FLIR #1 FLIR #2 FLIR #3 M3FD #1 M3FD #2 M3FD #3 [PITH_FULL_IMAGE:figures/full_fi… view at source ↗
Figure 6
Figure 6. Figure 6: Complexity distribution map. The red areas are allocated with more FLOPs, while the blue areas are allocated with the minimum computational cost. The appar￾ent backgrounds are filtered out, and the fusion is performed on the potential areas. 0.00 0.25 0.50 0.75 1.00 1.25 Area (M pixels) 0 100 200 300 FLOPS (G) FLOPS vs. Area Dense@18.8M Total@24.5M RPN@16.8M [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FPS vs. #GTs. SFEDet achieves compa￾rable efficiency on dense samples and much better on sparse samples. 0.0 2.5 5.0 7.5 10.0 Shifts in X-axis (pixels) 32 40 48 56 mAP EI2Det Ours Ours w/ Aug 0.0 2.5 5.0 7.5 10.0 Shifts in Y-axis (pixels) 32 40 48 56 mAP EI2Det Ours Ours w/ Aug [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
read the original abstract

RGB-T detectors leverage the complementary strengths of visible and thermal infrared modalities, achieving robust performance under challenging conditions. Many of them resort to heavy dual backbones and exhaustive cross-modality fusion across the entire image, leading to impractically high computational costs. We observe that most image regions are smooth backgrounds (e.g., sky, ground) that can be easily handled by lightweight single-modality models. In light of this observation, we propose a sparse fusion mechanism for efficient RGB-T detection: first rapidly scanning the image to identify the proposals and then carefully examining the remaining sparse proposals via feature fusion. We propose a two-stage framework to instantiate this mechanism, which performs detection in two stages: 1) a lightweight and modality-specific detection stage that produces high-recall RoIs, and 2) a fusion-driven examination and refinement stage that filters out the false positives and refines the bounding boxes. This design enables the detector to adaptively allocate more computational resources to the potential foregrounds, improving the efficiency while ensuring detection accuracy. Extensive experiments show that our method achieves competitive performance with substantially fewer parameters and lower cost, while maintaining strong scalability to high-resolution images.

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 / 1 minor

Summary. The manuscript proposes a two-stage sparse cross-modality fusion framework for efficient RGB-T object detection. A lightweight modality-specific first stage rapidly generates high-recall RoIs by exploiting the observation that most regions are smooth backgrounds; a second stage then performs feature fusion only on the sparse proposals to filter false positives and refine boxes. The authors claim this adaptively allocates compute to foregrounds, yielding competitive accuracy with substantially fewer parameters, lower cost, and good scalability to high-resolution images.

Significance. If the performance claims are substantiated, the work would provide a practical route to reducing the computational burden of dual-backbone RGB-T detectors by exploiting foreground sparsity, with relevance to real-time or resource-limited applications. The design choice is a direct response to the cost of exhaustive fusion and could influence subsequent efficient multi-modal architectures.

major comments (2)
  1. [Abstract] Abstract: the central claims of 'competitive performance with substantially fewer parameters and lower cost' are asserted without any quantitative tables, ablation studies, error bars, implementation details, or baseline comparisons, leaving the efficiency and accuracy assertions unsupported by visible evidence.
  2. [Abstract] Abstract: the load-bearing assumption that the lightweight single-modality first stage reliably produces high-recall RoIs (so that no object is irrecoverably missed before the fusion stage) is justified only by the qualitative 'smooth backgrounds' observation; no recall bounds, failure-mode analysis, or cross-condition validation is supplied, making the two-stage guarantee unverified.
minor comments (1)
  1. [Abstract] Abstract: the terms 'high-recall RoIs' and 'sparse proposals' are used without quantitative thresholds or definitions, which would help clarify the sparsity mechanism.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We agree that the abstract can be strengthened to better substantiate its claims and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claims of 'competitive performance with substantially fewer parameters and lower cost' are asserted without any quantitative tables, ablation studies, error bars, implementation details, or baseline comparisons, leaving the efficiency and accuracy assertions unsupported by visible evidence.

    Authors: The abstract is a concise summary, with the supporting quantitative evidence (including parameter counts, FLOPs, mAP comparisons to baselines, ablations, and implementation details) presented in the Experiments section of the full manuscript. We will revise the abstract to incorporate key quantitative highlights from those results to make the efficiency and accuracy claims more directly supported within the abstract itself. revision: yes

  2. Referee: [Abstract] Abstract: the load-bearing assumption that the lightweight single-modality first stage reliably produces high-recall RoIs (so that no object is irrecoverably missed before the fusion stage) is justified only by the qualitative 'smooth backgrounds' observation; no recall bounds, failure-mode analysis, or cross-condition validation is supplied, making the two-stage guarantee unverified.

    Authors: The abstract motivates the high-recall first stage via the smooth-background observation, with the full method and overall validation in later sections. To directly address the concern about verification of the two-stage guarantee, we will add a targeted analysis (including first-stage recall metrics, discussion of failure modes, and cross-condition checks) to the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity: architectural proposal with no derivation chain

full rationale

The paper advances a two-stage RGB-T detector design motivated by the observation that most image regions are smooth backgrounds. No equations, fitted parameters, predictions, or self-citations appear in the provided text. The high-recall RoI stage is presented as an empirical assumption enabling the sparse-fusion architecture rather than a quantity derived from or equivalent to the final result. The contribution is therefore a self-contained engineering choice validated externally by experiments, with no load-bearing step that reduces by construction to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The approach rests primarily on the domain assumption that backgrounds dominate images and can be filtered cheaply; no explicit free parameters or invented entities are named in the abstract.

axioms (1)
  • domain assumption Most image regions are smooth backgrounds that can be easily handled by lightweight single-modality models
    This observation is invoked to justify skipping fusion on the majority of the image.

pith-pipeline@v0.9.1-grok · 5740 in / 1241 out tokens · 27814 ms · 2026-06-30T06:14:48.938156+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

42 extracted references · 17 canonical work pages · 5 internal anchors

  1. [1]

    YOLOv4: Optimal Speed and Accuracy of Object Detection

    Bochkovskiy, A., Wang, C.Y., Liao, H.Y.M.: Yolov4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2004

  2. [2]

    Knowledge-Based Systems268, 110423 (May 2023).https://doi

    Chen, K., Liu, J., Zhang, H.: Igt: Illumination-guided rgb-t object detection with transformers. Knowledge-Based Systems268, 110423 (May 2023).https://doi. org/10.1016/j.knosys.2023.110423

    work page doi:10.1016/j.knosys.2023.110423 2023

  3. [3]

    In: European Conference on Computer Vision

    Chen, Y.T., Shi, J., Ye, Z., Mertz, C., Ramanan, D., Kong, S.: Multimodal ob- ject detection via probabilistic ensembling. In: European Conference on Computer Vision. pp. 139–158. Springer (2022)

    2022

  4. [4]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Deevi, S.A., Lee, C., Gan, L., Nagesh, S., Pandey, G., Chung, S.J.: Rgb-x ob- ject detection via scene-specific fusion modules. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 7366–7375 (2024)

    2024

  5. [6]

    IEEE Transactions on Multimedia p

    Dong, W., Zhu, H., Lin, S., Luo, X., Shen, Y., Guo, G., Zhang, B.: Fusion-mamba for cross-modality object detection. IEEE Transactions on Multimedia p. 1–15 (2025).https://doi.org/10.1109/TMM.2025.3599020

    work page doi:10.1109/tmm.2025.3599020 2025

  6. [7]

    In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV)

    Feng, C., Zhong, Y., Gao, Y., Scott, M.R., Huang, W.: Tood: Task-aligned one- stage object detection. In: 2021 IEEE/CVF International Conference on Computer Vision (ICCV). pp. 3490–3499. IEEE Computer Society (2021)

    2021

  7. [8]

    YOLOX: Exceeding YOLO Series in 2021

    Ge, Z., Liu, S., Wang, F., Li, Z., Sun, J.: Yolox: Exceeding yolo series in 2021. arXiv preprint arXiv:2107.08430 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  8. [9]

    In: First conference on language modeling (2024)

    Gu, A., Dao, T.: Mamba: Linear-time sequence modeling with selective state spaces. In: First conference on language modeling (2024)

    2024

  9. [10]

    In: Proceedings of the IEEE international conference on computer vision

    He, K., Gkioxari, G., Dollár, P., Girshick, R.: Mask r-cnn. In: Proceedings of the IEEE international conference on computer vision. pp. 2961–2969 (2017)

    2017

  10. [11]

    In: Proceedings of the 31st ACM International Conference on Multimedia

    He, X., Tang, C., Zou, X., Zhang, W.: Multispectral object detection via cross- modal conflict-aware learning. In: Proceedings of the 31st ACM International Conference on Multimedia. p. 1465–1474. ACM, Ottawa ON Canada (Oct 2023). https://doi.org/10.1145/3581783.3612651,https://dl.acm.org/doi/10. 1145/3581783.3612651

    work page doi:10.1145/3581783.3612651 2023

  11. [12]

    IEEE Transactions on Circuits and Systems for Video Technology35(7), 7101–7115 (July 2025).https://doi.org/10.1109/TCSVT.2025.3539625

    Hu, K., He, Y., Li, Y., Zhao, J., Chen, S., Kang, Y.: Ei²det: Edge-guided illumination-aware interactive learning for visible-infrared object detection. IEEE Transactions on Circuits and Systems for Video Technology35(7), 7101–7115 (July 2025).https://doi.org/10.1109/TCSVT.2025.3539625

    work page doi:10.1109/tcsvt.2025.3539625 2025

  12. [13]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Hwang, S., Park, J., Kim, N., Choi, Y., So Kweon, I.: Multispectral pedestrian detection: Benchmark dataset and baseline. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1037–1045 (2015)

    2015

  13. [14]

    In: Proceedings of the IEEE/CVF international con- ference on computer vision

    Jia, X., Zhu, C., Li, M., Tang, W., Zhou, W.: Llvip: A visible-infrared paired dataset for low-light vision. In: Proceedings of the IEEE/CVF international con- ference on computer vision. pp. 3496–3504 (2021)

    2021

  14. [15]

    IEEE Transactions on Circuits and Systems for Video Technology32(3), 1510–1523 (Mar 2022).https://doi.org/10.1109/ TCSVT.2021.3076466

    Kim, J.U., Park, S., Ro, Y.M.: Uncertainty-guided cross-modal learning for robust multispectral pedestrian detection. IEEE Transactions on Circuits and Systems for Video Technology32(3), 1510–1523 (Mar 2022).https://doi.org/10.1109/ TCSVT.2021.3076466

    work page arXiv 2022

  15. [16]

    Pattern Recognition Letters179, 144–150 (2024) Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion 17

    Lee, S., Park, J., Park, J.: Crossformer: Cross-guided attention for multi-modal object detection. Pattern Recognition Letters179, 144–150 (2024) Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion 17

    2024

  16. [17]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

    Li, F., Zhang, H., Liu, S., Guo, J., Ni, L.M., Zhang, L.: Dn-detr: Accelerate detr training by introducing query denoising. In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition. pp. 13619–13627 (2022)

    2022

  17. [18]

    IEEE Transactions on Multi- media (2022)

    Li, Q., Zhang, C., Hu, Q., Fu, H., Zhu, P.: Confidence-aware fusion using dempster- shafer theory for multispectral pedestrian detection. IEEE Transactions on Multi- media (2022)

    2022

  18. [19]

    Advances in neural information processing systems33, 21002–21012 (2020)

    Li, X., Wang, W., Wu, L., Chen, S., Hu, X., Li, J., Tang, J., Yang, J.: General- ized focal loss: Learning qualified and distributed bounding boxes for dense object detection. Advances in neural information processing systems33, 21002–21012 (2020)

    2020

  19. [20]

    Multispectral Deep Neural Networks for Pedestrian Detection

    Liu, J., Zhang, S., Wang, S., Metaxas, D.N.: Multispectral deep neural networks for pedestrian detection. arXiv preprint arXiv:1611.02644 (2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  20. [21]

    In: Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition

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

    2022

  21. [22]

    Cross-modality fusion transformer for multispectral object detection,

    Qingyun, F., Dapeng, H., Zhaokui, W.: Cross-modality fusion transformer for mul- tispectral object detection. arXiv preprint arXiv:2111.00273 (2021)

    work page arXiv 2021

  22. [23]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Redmon, J., Divvala, S., Girshick, R., Farhadi, A.: You only look once: Unified, real-time object detection. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 779–788 (2016)

    2016

  23. [24]

    YOLOv3: An Incremental Improvement

    Redmon, J., Farhadi, A.: Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  24. [25]

    Pattern Recog- nition145, 109913 (2024)

    Shen, J., Chen, Y., Liu, Y., Zuo, X., Fan, H., Yang, W.: Icafusion: Iterative cross- attention guided feature fusion for multispectral object detection. Pattern Recog- nition145, 109913 (2024)

    2024

  25. [26]

    Pattern Recognition Letters196, 1–8 (2025).https://doi.org/https://doi.org/10.1016/j.patrec.2025.05.005

    Tian, C., Yang, C., Zhu, G., Wang, Q., He, Z.: Learning a robust rgb-thermal detector for extreme modality imbalance. Pattern Recognition Letters196, 1–8 (2025).https://doi.org/https://doi.org/10.1016/j.patrec.2025.05.005

    work page doi:10.1016/j.patrec.2025.05.005 2025

  26. [27]

    IEEE Transactions on Multimedia26, 6449–6461 (2024)

    Tian, C., Zhou, Z., Huang, Y., Li, G., He, Z.: Cross-modality proposal-guided fea- ture mining for unregistered rgb-thermal pedestrian detection. IEEE Transactions on Multimedia26, 6449–6461 (2024)

    2024

  27. [28]

    Ultralytics:https://github.com/ultralytics/ultralytics

  28. [29]

    Computational Intelligence (2016)

    Wagner, J., Fischer, V., Herman, M., Behnke, S.: Multispectral pedestrian detec- tion using deep fusion convolutional neural networks. Computational Intelligence (2016)

    2016

  29. [30]

    Advances in Neural Information Processing Systems 37, 107984–108011 (2024)

    Wang, A., Chen, H., Liu, L., Chen, K., Lin, Z., Han, J., et al.: Yolov10: Real-time end-to-end object detection. Advances in Neural Information Processing Systems 37, 107984–108011 (2024)

    2024

  30. [31]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Wang, C.Y., Bochkovskiy, A., Liao, H.Y.M.: Yolov7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 7464–7475 (2023)

    2023

  31. [32]

    PP-YOLOE: An evolved version of YOLO,

    Xu, S., Wang, X., Lv, W., Chang, Q., Cui, C., Deng, K., Wang, G., Dang, Q., Wei, S., Du, Y., et al.: Pp-yoloe: An evolved version of yolo. arXiv preprint arXiv:2203.16250 (2022)

    work page arXiv 2022

  32. [34]

    IEEE Transactions on Circuits and Systems for Video Technology35(1), 547–560 (Jan 2025).https://doi.org/10.1109/TCSVT.2024

    Yang, F., Liang, B., Li, W., Zhang, J.: Multidimensional fusion network for mul- tispectral object detection. IEEE Transactions on Circuits and Systems for Video Technology35(1), 547–560 (Jan 2025).https://doi.org/10.1109/TCSVT.2024. 3454631

    work page doi:10.1109/tcsvt.2024 2025

  33. [35]

    DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection

    Zhang, H., Li, F., Liu, S., Zhang, L., Su, H., Zhu, J., Ni, L.M., Shum, H.Y.: 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 Pith/arXiv arXiv 2022

  34. [36]

    In: 2020 IEEE International conference on image processing (ICIP)

    Zhang, H., Fromont, E., Lefevre, S., Avignon, B.: Multispectral fusion for object detection with cyclic fuse-and-refine blocks. In: 2020 IEEE International conference on image processing (ICIP). pp. 276–280. IEEE (2020)

    2020

  35. [37]

    In: Proceedings of the IEEE/CVF winter conference on applications of computer vision

    Zhang, H., Fromont, E., Lefèvre, S., Avignon, B.: Guided attentive feature fusion for multispectral pedestrian detection. In: Proceedings of the IEEE/CVF winter conference on applications of computer vision. pp. 72–80 (2021)

    2021

  36. [38]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Zhang, L., Zhu, X., Chen, X., Yang, X., Lei, Z., Liu, Z.: Weakly aligned cross-modal learning for multispectral pedestrian detection. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 5127–5137 (2019)

    2019

  37. [39]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Zhang, X., Zhou, X., Lin, M., Sun, J.: Shufflenet: An extremely efficient convolu- tional neural network for mobile devices. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 6848–6856 (2018)

    2018

  38. [40]

    & Yin, J

    Zhang, X., Cao, S.Y., Wang, F., Zhang, R., Wu, Z., Zhang, X., Bai, X., Shen, H.L.: Rethinking early-fusion strategies for improved multispectral object detec- tion. IEEE Transactions on Intelligent Vehicles p. 1–15 (2024).https://doi.org/ 10.1109/TIV.2024.3462488

    work page doi:10.1109/tiv.2024.3462488 2024

  39. [41]

    In: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    Zhao, Y., Lv, W., Xu, S., Wei, J., Wang, G., Dang, Q., Liu, Y., Chen, J.: Detrs beat yolos on real-time object detection. In: 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). p. 16965–16974. IEEE, Seattle, WA, USA (june 2024).https://doi.org/10.1109/CVPR52733.2024.01605,https: //ieeexplore.ieee.org/document/10657220/

    work page doi:10.1109/cvpr52733.2024.01605 2024

  40. [42]

    In: European conference on computer vi- sion

    Zhou, K., Chen, L., Cao, X.: Improving multispectral pedestrian detection by ad- dressing modality imbalance problems. In: European conference on computer vi- sion. pp. 787–803. Springer (2020)

    2020

  41. [43]

    IEEE TransactionsonMultimediap.1–15(2025).https://doi.org/10.1109/TMM.2025

    Zhou, M., Li, Y., Yang, G., Wei, X., Pu, H., Luo, J., Jia, W.: Cofnet: Contrastive object-aware fusion using box-level masks for multispectral object detection. IEEE TransactionsonMultimediap.1–15(2025).https://doi.org/10.1109/TMM.2025. 3599097

    work page doi:10.1109/tmm.2025 2025

  42. [44]

    Zhu, X., Su, W., Lu, L., Li, B., Wang, X., Dai, J.: Deformable detr: Deformable transformers for end-to-end object detection. In: International Conference on Learning Representations (2021) Efficient RGB-T Object Detection via Sparse Cross-Modality Fusion 1 Appendix A Effect of the proposed FER We conduct a comparison to evaluate the effectiveness of the ...

    2021