Pith. sign in

REVIEW 3 major objections 5 minor 58 references

Fog mainly hurts UAV detection and tracking by causing missed detections, and training on foggy data beats test-time dehazing for robustness.

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

2026-07-11 14:26 UTC pith:GNRQSJHG

load-bearing objection Solid empirical ranking paper: fog-inclusive training beats test-time dehazing for small-UAV detection/tracking under controlled synthetic fog, with public multi-severity datasets. the 3 major comments →

arxiv 2607.05467 v1 pith:GNRQSJHG submitted 2026-07-06 cs.CV cs.LGeess.IV

A Task-Driven Evaluation of UAV Detection and Tracking under Synthetic Fog

Amir Pouladi , Vesal Ahsani , Haijun Li , Homayoun Najjaran , Afzal Suleman This is my paper
classification cs.CV cs.LGeess.IV
keywords UAV detectionsynthetic fogimage dehazingtracking-by-detectionadverse weathersense and avoidatmospheric scattering modeldepth estimation
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.

This paper builds a controlled pipeline that adds depth-aware synthetic fog to real clear-sky UAV images and videos, then measures how fog, image dehazing, and fog-aware training affect detection and tracking of small distant drones. Fog’s main damage is more missed targets, not a flood of false alarms. Training detectors on mixtures of clean and foggy images is the most reliable way to keep performance stable across fog densities, while dehazing at test time helps mainly when the detector never saw fog during training. Pretty restored images do not automatically mean better detection or tracking, so restoration must be judged together with the perception tasks that actually matter for sense-and-avoid.

Core claim

Fog substantially degrades detection and tracking of small UAVs in sky-dominant long-range imagery, primarily through increased missed detections. Fog-inclusive detector training provides the most consistent robustness gains across severities, whereas test-time restoration is most beneficial when the detector was trained only on clean imagery. Image-level restoration quality does not necessarily produce proportional gains in detection or tracking metrics, so restoration should be evaluated jointly with downstream perception performance.

What carries the argument

A depth-aware synthetic fog pipeline that estimates monocular relative depth, normalizes it by split-level percentiles, and applies the atmospheric scattering model with controllable severity β to produce paired clear/foggy UAV images and videos that feed a unified evaluation of restoration, clean-only versus fog-inclusive detector training, and tracking-by-detection.

Load-bearing premise

The results rest on synthetic fog from monocular relative depth and the atmospheric scattering model, with hand-chosen density levels, behaving enough like real outdoor fog that the ranking of training regimes and restoration benefits will transfer.

What would settle it

Re-run the same clean-only versus fog-inclusive detectors, with and without the selected dehazer, on a large set of real foggy anti-UAV videos with ground-truth boxes; if the fog-inclusive training advantage disappears or reverses while test-time restoration becomes dominant, the central claim fails.

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

If this is right

  • Detectors trained only on clear weather will suffer large recall drops under fog and need either fog-inclusive retraining or test-time dehazing.
  • Once a detector is trained with foggy samples, the extra value of running a dehazer before detection shrinks and can reverse.
  • Tracking-by-detection under fog is limited mainly by detector false negatives; tracker choice is secondary.
  • Practical systems may need condition-aware switching of restoration or detector weights based on estimated visibility.
  • Restoration can still aid human operators even when automated detection and tracking gains are small.

Where Pith is reading between the lines

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

  • The same ranking of fog-inclusive training over pure restoration is likely to hold for rain, low light, and other depth-dependent degradations on small aerial targets.
  • Edge-deployed anti-UAV systems may prefer lightweight fog-augmented training over a separate heavy dehazer to meet real-time budgets.
  • Public multi-severity synthetic foggy UAV sets could become a standard stress test for sense-and-avoid detectors, analogous to synthetic fog benchmarks in driving.
  • Multimodal fusion such as camera–radar may be required once synthetic fog alone no longer closes the gap to real adverse weather.

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

3 major / 5 minor

Summary. This paper presents a controlled, task-driven evaluation pipeline for small-UAV detection and tracking under synthetic fog. Using MiDaS monocular depth and the atmospheric scattering model, the authors generate multi-severity foggy variants of real clear-weather UAV imagery, compare classical/CNN/transformer dehazers, and then measure YOLO11 detection under clean-only versus fog-inclusive training and ByteTrack/BoT-SORT tracking-by-detection on clean, foggy, and restored sequences. The central empirical claim is that fog mainly increases missed detections; fog-inclusive training is the most consistent robustness lever; test-time restoration helps primarily clean-trained detectors; and image-level restoration metrics do not guarantee proportional gains in detection or tracking.

Significance. If the reported rankings hold under the stated synthetic protocol, the work is a useful systems-level contribution for anti-UAV and sense-and-avoid perception: it cleanly separates training-time robustness from test-time restoration and shows that dehazing should not be evaluated in isolation for small, sky-dominant targets. Strengths include a unified clean/foggy/restored protocol on identical scenes, multiple detector scales and fog-mix ratios, dual trackers, standard detection/tracking metrics, public release of the synthetic foggy datasets, and an explicit limitation statement on real-fog transfer. The study is primarily empirical rather than algorithmic novelty, but the controlled ranking is actionable for practitioners choosing between fog-aware training and restoration preprocessing.

major comments (3)
  1. [§3.3 / Table 1] §3.3 and Table 1: DehazeFormer is selected solely from MMAUD PSNR/SSIM, then applied unchanged to CfAR detection and DUT tracking without reporting restoration metrics or qualitative checks on those target datasets. Because the central claim compares restored vs foggy downstream performance, the paper should either report PSNR/SSIM (or at least failure cases) on CfAR/DUT or justify that MMAUD ranking transfers to those domains.
  2. [§4.3 / Tables 3–4] §4.3 and Tables 3–4: Tracking conclusions rest on four labeled DUT sample videos (02, 03, 06, 10). Sequence difficulty clearly dominates (Video 10 near-collapse; Video 06 stable), so the claim that fog-aware training transfers to stronger tracking-by-detection is only weakly supported. Expand to a larger DUT subset with aggregate MOTA/IDF1, or clearly reframe tracking as illustrative case studies rather than a general result.
  3. [§3.2 / Eqs. (1)–(2)] §3.2, Eqs. (1)–(2): The load-bearing proxy assumption—that MiDaS relative depth with fixed split-level percentile normalization and hand-chosen β ∈ {0.4,…,3.6} ranks training/restoration regimes as real sky-dominant fog would—is acknowledged in §5–§6 but not stress-tested. At minimum, add a sensitivity analysis (alternate depth estimators, alternate A estimation, or β rescaling) showing that the clean-only vs fog-inclusive ranking is stable under reasonable synthesis choices.
minor comments (5)
  1. [Title / Abstract] Title and abstract: fix spacing typos such as “UA V” and “skydominant”.
  2. [Figures 4–7] Figures 4–7 are dense multi-panel plots; axis labels and legend contrast make the clean baseline hard to read in grayscale. Consider larger fonts or separate clean-trained panels.
  3. [§3.4] §3.4: state the random seed / replacement protocol for fog-mix ratios more explicitly so the 30/50/70/100% regimes are fully reproducible.
  4. [§2] Related work is thorough but long relative to the experimental core; a short table contrasting prior UAV fog datasets (FDD, VisDrone Foggy, HazyDet) with the present multi-β, depth-normalized protocol would help readers place the contribution.
  5. [§4.1] Keywords and abbreviations are fine; ensure “mAP@0.50:0.95” is defined once before first use in the results narrative (it is defined later in §4.1).

Circularity Check

0 steps flagged

No circular derivation: empirical rankings under a controlled synthetic protocol, not predictions forced by fitted inputs or self-citation.

full rationale

This is a task-driven empirical evaluation paper, not a first-principles derivation. The load-bearing claims (fog mainly increases missed detections; fog-inclusive training is the most consistent robustness lever; test-time restoration helps clean-trained detectors most; image-level PSNR/SSIM gains need not track mAP/MOTA/IDF1) are measured outcomes on held-out clean/foggy/restored splits of CfAR and DUT Anti-UAV, using YOLO11 variants and ByteTrack/BoT-SORT (Figs. 4–9, Tables 3–4). Synthetic fog is generated by an external monocular depth model (MiDaS) plus the standard atmospheric scattering model with free experimental choices of β ∈ {0.4, 0.8, 1.6, 2.4, 3.6} and fog-mix ratios; those parameters are not fitted to the reported detection/tracking metrics and then re-presented as predictions. Restoration model selection (DehazeFormer) is an upstream comparison on MMAUD, not a circular definition of the downstream result. Self-citations are to public datasets used as inputs (DUT, MMAUD, CfAR), not to uniqueness theorems or load-bearing prior results that force the claimed ranking. No step reduces by construction to its own inputs; the only acknowledged external assumption is synthetic-to-real transfer, which is a validity limit, not circularity.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central claims rest on standard atmospheric image formation, monocular depth as a usable proxy, and the experimental choices of fog densities and training mix ratios. No new physical entities are postulated; free parameters are the controllable experimental knobs of the synthesis and training regimes.

free parameters (4)
  • fog severity β = {0.4, 0.8, 1.6, 2.4, 3.6}
    Hand-chosen discrete set {0.4, 0.8, 1.6, 2.4, 3.6} that defines all foggy test conditions and the severity axis of every plot; not fitted to real fog measurements.
  • fog-inclusive training mix ratios = 0/30/50/70/100 %
    Clean / 30% / 50% / 70% / 100% fog replacement fractions that define the training regimes whose robustness is compared.
  • depth percentile bounds p_low, p_high
    Split-level percentiles used to normalize MiDaS relative depth before inversion; choice affects transmission maps and therefore all synthetic fog.
  • atmospheric light A estimation
    Estimated from brightest pixels of each input image; standard heuristic but still a free implementation choice that modulates fog appearance.
axioms (4)
  • domain assumption Atmospheric scattering model I = J t + A(1-t) with t = exp(-β d) adequately models fog degradation for sky-dominant UAV imagery.
    Invoked throughout §3.2 and used to generate every foggy image/video; standard in the literature but unvalidated against real fog in this paper.
  • domain assumption MiDaS monocular relative depth, after percentile normalization and inversion, is a sufficiently accurate depth map for controllable fog synthesis.
    Core of the synthesis pipeline in §3.2; scale ambiguity is acknowledged but treated as acceptable for severity control.
  • ad hoc to paper DehazeFormer selected on MMAUD restoration metrics is the appropriate restoration model for all downstream CfAR detection and DUT tracking experiments.
    Selection in §3.3 / Table 1; no ablation of alternative restorers inside the detection/tracking loops.
  • standard math Standard detection (mAP, P, R) and tracking (MOTA, IDF1) metrics at the reported IoU/association settings correctly rank robustness for the anti-UAV task.
    Used throughout §4; conventional definitions from the MOT and detection literature.

pith-pipeline@v1.1.0-grok45 · 27546 in / 3118 out tokens · 28205 ms · 2026-07-11T14:26:47.553697+00:00 · methodology

0 comments
read the original abstract

Fog severely degrades the visibility of small unmanned aerial vehicles (UAVs) in skydominant, long-range imagery, reducing the reliability of downstream detection and tracking. This paper presents a task-driven evaluation framework that links depth-aware synthetic fog generation, image restoration, object detection, and tracking within a unified pipeline. Given the practical difficulty of collecting and annotating foggy UAV scenes, synthetic fog is generated from real clear-weather outdoor images containing UAV targets using monocular depth estimation and the atmospheric scattering model. Representative restoration methods from classical, convolutional neural network (CNN)-based, and transformer-based families are first compared, after which the selected restoration model is integrated into the downstream perception pipeline. Detection is evaluated under both clean-only and fog-inclusive training regimes using multiple detector variants, while tracking-by-detection is assessed on clean, foggy, and restored video sequences. Beyond image-level restoration metrics, the study evaluates how fog and restoration affect detection robustness and tracking performance. The results show that fog substantially degrades both detection and tracking, primarily through increased missed detections. Fog-inclusive training provides the most consistent improvement in robustness, whereas test-time restoration is most beneficial when the detector has been trained only on clean imagery. These findings show that restoration quality does not necessarily translate into proportional gains in downstream perception and therefore should be evaluated jointly with detection and tracking performance.

Figures

Figures reproduced from arXiv: 2607.05467 by Afzal Suleman, Amir Pouladi, Haijun Li, Homayoun Najjaran, Vesal Ahsani.

Figure 1
Figure 1. Figure 1: End-to-end experimental pipeline used in this work. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Depth-aware synthetic fog generation used in this study. Clean images are processed [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison of restoration methods across fog severities over MMAUD [ [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Precision (P) of YOLO11n, YOLO11s, and YOLO11m under different training￾set compositions and evaluation conditions. Rows correspond to detector variants, columns correspond to training configurations, while the curves show performance on foggy and restored test sets across increasing fog severity. The dashed line denotes performance on the clean test set. Second, mAP@0.50 and mAP@0.50:0.95 closely follow t… view at source ↗
Figure 5
Figure 5. Figure 5: Recall (R) of the evaluated YOLO11 variants, under the same training and evaluation setup as in [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Detection performance in terms of mAP@0.50 for the YOLO11 variants, under the [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Detection performance in terms of mAP@0.50:0.95 for the YOLO11 variants, under [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Representative qualitative examples showing cases in which the YOLO11n detector [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Detection performance and stability of YOLO11n, YOLO11s, and YOLO11m trained [PITH_FULL_IMAGE:figures/full_fig_p016_9.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

58 extracted references · 4 linked inside Pith

  1. [1]

    Optics of the atmosphere: scattering by molecules and particles.New york, 1976

    Earl J McCartney. Optics of the atmosphere: scattering by molecules and particles.New york, 1976

    1976

  2. [2]

    Vision transformers for single image dehazing.IEEE Transactions on Image Processing, 32:1927–1941, 2023

    Yuda Song, Zhuqing He, Hui Qian, and Xin Du. Vision transformers for single image dehazing.IEEE Transactions on Image Processing, 32:1927–1941, 2023

    1927

  3. [3]

    Benchmarking single-image dehazing and beyond.IEEE transactions on image processing, 28(1):492–505, 2018

    Boyi Li, Wenqi Ren, Dengpan Fu, Dacheng Tao, Dan Feng, Wenjun Zeng, and Zhangyang Wang. Benchmarking single-image dehazing and beyond.IEEE transactions on image processing, 28(1):492–505, 2018

    2018

  4. [4]

    Anti-uav: A large multi-modal benchmark for uav tracking.arXiv preprint arXiv:2101.08466, 2021

    Nan Jiang, Kuiran Wang, Xiaoke Peng, Xuehui Yu, Qiang Wang, Junliang Xing, Guorong Li, Jian Zhao, Guodong Guo, and Zhenjun Han. Anti-uav: A large multi-modal benchmark for uav tracking.arXiv preprint arXiv:2101.08466, 2021

    Pith/arXiv arXiv 2021

  5. [5]

    Vision-based anti-uav detection and tracking.IEEE Transactions on Intelligent Transportation Systems, 23(12):25323–25334, 2022

    Jie Zhao, Jingshu Zhang, Dongdong Li, and Dong Wang. Vision-based anti-uav detection and tracking.IEEE Transactions on Intelligent Transportation Systems, 23(12):25323–25334, 2022

    2022

  6. [6]

    Mmaud: A comprehensive multi-modal anti-uav dataset for modern miniature drone threats

    Shenghai Yuan, Yizhuo Yang, Thien Hoang Nguyen, Thien-Minh Nguyen, Jianfei Yang, Fen Liu, Jianping Li, Han Wang, and Lihua Xie. Mmaud: A comprehensive multi-modal anti-uav dataset for modern miniature drone threats. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 2745–2751. IEEE, 2024

    2024

  7. [7]

    Vision and the atmosphere.International journal of computer vision, 48(3):233–254, 2002

    Srinivasa G Narasimhan and Shree K Nayar. Vision and the atmosphere.International journal of computer vision, 48(3):233–254, 2002

    2002

  8. [8]

    Semantic foggy scene understanding with synthetic data.International Journal of Computer Vision, 126(9):973–992, 2018

    Christos Sakaridis, Dengxin Dai, and Luc Van Gool. Semantic foggy scene understanding with synthetic data.International Journal of Computer Vision, 126(9):973–992, 2018

    2018

  9. [9]

    Model adaptation with synthetic and real data for semantic dense foggy scene understanding

    Christos Sakaridis, Dengxin Dai, Simon Hecker, and Luc Van Gool. Model adaptation with synthetic and real data for semantic dense foggy scene understanding. InProceedings of the european conference on computer vision (ECCV), pages 687–704, 2018

    2018

  10. [10]

    Foggy lane dataset synthesized from monocular images for lane detection algorithms.Sensors, 22(14): 5210, 2022

    Xiangyu Nie, Zhejun Xu, Wei Zhang, Xue Dong, Ning Liu, and Yuanfeng Chen. Foggy lane dataset synthesized from monocular images for lane detection algorithms.Sensors, 22(14): 5210, 2022

    2022

  11. [11]

    A foggy weather simulation algorithm for traffic image synthesis based on monocular depth estimation.Sensors, 24(6):1966, 2024

    Minan Tang, Zixin Zhao, and Jiandong Qiu. A foggy weather simulation algorithm for traffic image synthesis based on monocular depth estimation.Sensors, 24(6):1966, 2024

    1966

  12. [12]

    Synfog: A photo- realistic synthetic fog dataset based on end-to-end imaging simulation for advancing real- world defogging in autonomous driving

    Yiming Xie, Henglu Wei, Zhenyi Liu, Xiaoyu Wang, and Xiangyang Ji. Synfog: A photo- realistic synthetic fog dataset based on end-to-end imaging simulation for advancing real- world defogging in autonomous driving. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 21763–21772, 2024

    2024

  13. [13]

    Fog simulation on real lidar point clouds for 3d object detection in adverse weather

    Martin Hahner, Christos Sakaridis, Dengxin Dai, and Luc Van Gool. Fog simulation on real lidar point clouds for 3d object detection in adverse weather. InProceedings of the IEEE/CVF international conference on computer vision, pages 15283–15292, 2021

    2021

  14. [14]

    Foggy drone teacher: Domain adaptive drone detection under foggy conditions.Drones, 9(2):146, 2025

    Guida Zheng, Benying Tan, Jingxin Wu, Xiao Qin, Yujie Li, and Shuxue Ding. Foggy drone teacher: Domain adaptive drone detection under foggy conditions.Drones, 9(2):146, 2025

    2025

  15. [15]

    Fifonet: fine-grained target focusing network for object detection in uav images.Remote Sensing, 14 (16):3919, 2022

    Yue Xi, Wenjing Jia, Qiguang Miao, Xiangzeng Liu, Xiaochen Fan, and Hanhui Li. Fifonet: fine-grained target focusing network for object detection in uav images.Remote Sensing, 14 (16):3919, 2022. 22

    2022

  16. [16]

    Detection, tracking, and counting meets drones in crowds: A benchmark

    Longyin Wen, Dawei Du, Pengfei Zhu, Qinghua Hu, Qilong Wang, Liefeng Bo, and Siwei Lyu. Detection, tracking, and counting meets drones in crowds: A benchmark. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 7812–7821, 2021

    2021

  17. [17]

    Multi-task learning for uav aerial object detection in foggy weather condition.Remote Sensing, 15(18):4617, 2023

    Wenxuan Fang, Guoqing Zhang, Yuhui Zheng, and Yuwen Chen. Multi-task learning for uav aerial object detection in foggy weather condition.Remote Sensing, 15(18):4617, 2023

    2023

  18. [18]

    Hazydet: Open-source benchmark for drone-view object detection with depth-cues in hazy scenes.arXiv preprint arXiv:2409.19833, 2024

    Changfeng Feng, Zhenyuan Chen, Xiang Li, Chunping Wang, Jian Yang, Ming-Ming Cheng, Yimian Dai, and Qiang Fu. Hazydet: Open-source benchmark for drone-view object detection with depth-cues in hazy scenes.arXiv preprint arXiv:2409.19833, 2024

    Pith/arXiv arXiv 2024

  19. [19]

    Aerial autonomy under adversity: Advances in obstacle and aircraft detection techniques for unmanned aerial vehicles.Drones, 9(8):549, 2025

    Cristian Randieri, Sai Venkata Ganesh, Rayappa David Amar Raj, Rama Muni Reddy Yanamala, Archana Pallakonda, and Christian Napoli. Aerial autonomy under adversity: Advances in obstacle and aircraft detection techniques for unmanned aerial vehicles.Drones, 9(8):549, 2025

    2025

  20. [20]

    Visibility in bad weather from a single image

    Robby T Tan. Visibility in bad weather from a single image. In2008 IEEE conference on computer vision and pattern recognition, pages 1–8. IEEE, 2008

    2008

  21. [21]

    Single image dehazing.ACM transactions on graphics (TOG), 27(3):1–9, 2008

    Raanan Fattal. Single image dehazing.ACM transactions on graphics (TOG), 27(3):1–9, 2008

    2008

  22. [22]

    Single image haze removal using dark channel prior.IEEE transactions on pattern analysis and machine intelligence, 33(12):2341–2353, 2010

    Kaiming He, Jian Sun, and Xiaoou Tang. Single image haze removal using dark channel prior.IEEE transactions on pattern analysis and machine intelligence, 33(12):2341–2353, 2010

    2010

  23. [23]

    Efficient image dehazing with boundary constraint and contextual regularization

    Gaofeng Meng, Ying Wang, Jiangyong Duan, Shiming Xiang, and Chunhong Pan. Efficient image dehazing with boundary constraint and contextual regularization. InProceedings of the IEEE international conference on computer vision, pages 617–624, 2013

    2013

  24. [24]

    A fast single image haze removal algorithm using color attenuation prior.IEEE transactions on image processing, 24(11):3522–3533, 2015

    Qingsong Zhu, Jiaming Mai, and Ling Shao. A fast single image haze removal algorithm using color attenuation prior.IEEE transactions on image processing, 24(11):3522–3533, 2015

    2015

  25. [25]

    Non-local image dehazing

    Dana Berman, Shai Avidan, et al. Non-local image dehazing. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1674–1682, 2016

    2016

  26. [26]

    Single image dehazing using haze-lines

    Dana Berman, Tali Treibitz, and Shai Avidan. Single image dehazing using haze-lines. IEEE transactions on pattern analysis and machine intelligence, 42(3):720–734, 2018

    2018

  27. [27]

    Dehazenet: An end-to-end system for single image haze removal.IEEE transactions on image processing, 25(11):5187–5198, 2016

    Bolun Cai, Xiangmin Xu, Kui Jia, Chunmei Qing, and Dacheng Tao. Dehazenet: An end-to-end system for single image haze removal.IEEE transactions on image processing, 25(11):5187–5198, 2016

    2016

  28. [28]

    Single image dehazing via multi-scale convolutional neural networks

    Wenqi Ren, Si Liu, Hua Zhang, Jinshan Pan, Xiaochun Cao, and Ming-Hsuan Yang. Single image dehazing via multi-scale convolutional neural networks. InEuropean conference on computer vision, pages 154–169. Springer, 2016

    2016

  29. [29]

    Aod-net: All-in-one dehazing network

    Boyi Li, Xiulian Peng, Zhangyang Wang, Jizheng Xu, and Dan Feng. Aod-net: All-in-one dehazing network. InProceedings of the IEEE international conference on computer vision, pages 4770–4778, 2017

    2017

  30. [30]

    Griddehazenet: Attention-based multi-scale network for image dehazing

    Xiaohong Liu, Yongrui Ma, Zhihao Shi, and Jun Chen. Griddehazenet: Attention-based multi-scale network for image dehazing. InProceedings of the IEEE/CVF international conference on computer vision, pages 7314–7323, 2019. 23

    2019

  31. [31]

    Multi-scale boosted dehazing network with dense feature fusion

    Hang Dong, Jinshan Pan, Lei Xiang, Zhe Hu, Xinyi Zhang, Fei Wang, and Ming-Hsuan Yang. Multi-scale boosted dehazing network with dense feature fusion. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 2157–2167, 2020

    2020

  32. [32]

    Ffa-net: Feature fusion attention network for single image dehazing

    Xu Qin, Zhilin Wang, Yuanchao Bai, Xiaodong Xie, and Huizhu Jia. Ffa-net: Feature fusion attention network for single image dehazing. InProceedings of the AAAI conference on artificial intelligence, volume 34, pages 11908–11915, 2020

    2020

  33. [33]

    Contrastive learning for compact single image dehazing

    Haiyan Wu, Yanyun Qu, Shaohui Lin, Jian Zhou, Ruizhi Qiao, Zhizhong Zhang, Yuan Xie, and Lizhuang Ma. Contrastive learning for compact single image dehazing. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10551–10560, 2021

    2021

  34. [34]

    Curricular contrastive regularization for physics-aware single image dehazing

    Yu Zheng, Jiahui Zhan, Shengfeng He, Junyu Dong, and Yong Du. Curricular contrastive regularization for physics-aware single image dehazing. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5785–5794, 2023

    2023

  35. [35]

    Ridcp: Revitalizing real image dehazing via high-quality codebook priors

    Rui-Qi Wu, Zheng-Peng Duan, Chun-Le Guo, Zhi Chai, and Chongyi Li. Ridcp: Revitalizing real image dehazing via high-quality codebook priors. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22282–22291, 2023

    2023

  36. [36]

    Dea-net: Single image dehazing based on detail-enhanced convolution and content-guided attention.IEEE transactions on image processing, 33:1002–1015, 2024

    Zixuan Chen, Zewei He, and Zhe-Ming Lu. Dea-net: Single image dehazing based on detail-enhanced convolution and content-guided attention.IEEE transactions on image processing, 33:1002–1015, 2024

    2024

  37. [37]

    Image dehazing transformer with transmission-aware 3d position embedding

    Chun-Le Guo, Qixin Yan, Saeed Anwar, Runmin Cong, Wenqi Ren, and Chongyi Li. Image dehazing transformer with transmission-aware 3d position embedding. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5812–5820, 2022

    2022

  38. [38]

    Interaction-guided two-branch image dehazing network

    Huichun Liu, Xiaosong Li, and Tianshu Tan. Interaction-guided two-branch image dehazing network. InProceedings of the Asian conference on computer vision, pages 4069–4084, 2024

    2024

  39. [39]

    High-quality image dehazing with diffusion model.arXiv preprint arXiv:2308.11949, 2023

    Hu Yu, Jie Huang, Kaiwen Zheng, and Feng Zhao. High-quality image dehazing with diffusion model.arXiv preprint arXiv:2308.11949, 2023

    Pith/arXiv arXiv 2023

  40. [40]

    Unleashing the potential of the semantic latent space in diffusion models for image dehazing

    Zizheng Yang, Hu Yu, Bing Li, Jinghao Zhang, Jie Huang, and Feng Zhao. Unleashing the potential of the semantic latent space in diffusion models for image dehazing. InEuropean Conference on Computer Vision, pages 371–389. Springer, 2024

    2024

  41. [41]

    Efficient image dehazing via temporal-aware diffusion.Expert Systems with Applications, 291:128565, 2025

    Haobo Liang, Yan Yang, and Jiajie Jing. Efficient image dehazing via temporal-aware diffusion.Expert Systems with Applications, 291:128565, 2025

    2025

  42. [42]

    Latent space diffusion model for image dehazing

    Haobo Liang, Yan Yang, and Jiajie Jing. Latent space diffusion model for image dehazing. Applied Soft Computing, 180:113322, 2025

    2025

  43. [43]

    Robust object detection in challenging weather conditions

    Himanshu Gupta, Oleksandr Kotlyar, Henrik Andreasson, and Achim J Lilienthal. Robust object detection in challenging weather conditions. InProceedings of the IEEE/CVF winter conference on applications of computer vision, pages 7523–7532, 2024

    2024

  44. [44]

    Auxiliary domain- guided adaptive detection in adverse weather conditions

    Zhuobin Fu, Kan Chang, Mingyang Ling, Qingzhi Zhang, and Enze Qi. Auxiliary domain- guided adaptive detection in adverse weather conditions. InProceedings of the Asian Conference on Computer Vision, pages 3964–3981, 2024

    2024

  45. [45]

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

    Yuka Ogino, Yuho Shoji, Takahiro Toizumi, and Atsushi Ito. Erup-yolo: Enhancing object detection robustness for adverse weather condition by unified image-adaptive processing. In 2025 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 8597–8605. IEEE, 2025. 24

    2025

  46. [46]

    Evaluation of spatio-temporal small object detection in real-world adverse weather conditions

    Michel Van Lier, Martin Van Leeuwen, Bastian Van Manen, Leo Kampmeijer, and Nicolas Boehrer. Evaluation of spatio-temporal small object detection in real-world adverse weather conditions. InProceedings of the Winter Conference on Applications of Computer Vision, pages 844–855, 2025

    2025

  47. [47]

    A two-stage method for aerial tracking in adverse weather conditions.Mathematics, 12(8):1216, 2024

    Yuan Feng, Xinnan Xu, Nuoyi Chen, Quanjian Song, and Lufang Zhang. A two-stage method for aerial tracking in adverse weather conditions.Mathematics, 12(8):1216, 2024

    2024

  48. [48]

    Improved deepsort-based object tracking in foggy weather for avs using sematic labels and fused appearance feature network.Sensors, 24(14): 4692, 2024

    Isaac Ogunrinde and Shonda Bernadin. Improved deepsort-based object tracking in foggy weather for avs using sematic labels and fused appearance feature network.Sensors, 24(14): 4692, 2024

    2024

  49. [49]

    Seeing through fog without seeing fog: Deep multimodal sensor fusion in unseen adverse weather

    Mario Bijelic, Tobias Gruber, Fahim Mannan, Florian Kraus, Werner Ritter, Klaus Diet- mayer, and Felix Heide. Seeing through fog without seeing fog: Deep multimodal sensor fusion in unseen adverse weather. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11682–11692, 2020

    2020

  50. [50]

    Vision transformers for dense prediction

    Ren´ e Ranftl, Alexey Bochkovskiy, and Vladlen Koltun. Vision transformers for dense prediction. InProceedings of the IEEE/CVF international conference on computer vision, pages 12179–12188, 2021

    2021

  51. [51]

    Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing, 13(4):600–612, 2004

    2004

  52. [52]

    Ultralytics yolo11

    Glenn Jocher and Jing Qiu. Ultralytics yolo11. https://github.com/ultralytics/ ultralytics, 2024. Version 11.0.0

    2024

  53. [53]

    Infrared and visible camera integration for detection and tracking of small uavs: Systematic evaluation

    Ana Pereira, Stephen Warwick, Alexandra Moutinho, and Afzal Suleman. Infrared and visible camera integration for detection and tracking of small uavs: Systematic evaluation. Drones, 8(11):650, 2024

    2024

  54. [54]

    Bytetrack: Multi-object tracking by associating every detection box

    Yifu Zhang, Peize Sun, Yi Jiang, Dongdong Yu, Fucheng Weng, Zehuan Yuan, Ping Luo, Wenyu Liu, and Xinggang Wang. Bytetrack: Multi-object tracking by associating every detection box. InEuropean conference on computer vision, pages 1–21. Springer, 2022

    2022

  55. [55]

    Bot-sort: Robust associations multi- pedestrian tracking.arXiv preprint arXiv:2206.14651, 2022

    Nir Aharon, Roy Orfaig, and Ben-Zion Bobrovsky. Bot-sort: Robust associations multi- pedestrian tracking.arXiv preprint arXiv:2206.14651, 2022

    Pith/arXiv arXiv 2022

  56. [56]

    The pascal visual object classes (voc) challenge.International journal of computer vision, 88(2):303–338, 2010

    Mark Everingham, Luc Van Gool, Christopher KI Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge.International journal of computer vision, 88(2):303–338, 2010

    2010

  57. [57]

    Evaluating multiple object tracking performance: the clear mot metrics.EURASIP Journal on Image and Video Processing, 2008(1):246309, 2008

    Keni Bernardin and Rainer Stiefelhagen. Evaluating multiple object tracking performance: the clear mot metrics.EURASIP Journal on Image and Video Processing, 2008(1):246309, 2008

    2008

  58. [58]

    Performance measures and a data set for multi-target, multi-camera tracking

    Ergys Ristani, Francesco Solera, Roger Zou, Rita Cucchiara, and Carlo Tomasi. Performance measures and a data set for multi-target, multi-camera tracking. InEuropean conference on computer vision, pages 17–35. Springer, 2016. 25

    2016