pith. sign in

arxiv: 2606.26828 · v1 · pith:YBV56HFMnew · submitted 2026-06-25 · 💻 cs.CV

Learning Adversarial Augmentation Policies for Robust Garlic Seedling Detection

Soeun Lee , Chanho Kim , Yeji Kang , YoungKi Hong , Byeongkeun Kang This is my paper

Pith reviewed 2026-06-26 05:46 UTC · model grok-4.3

classification 💻 cs.CV
keywords garlic seedling detectionadversarial augmentation policyillumination robustnessprecision agricultureobject detectionmissing seedling localizationground-based monitoring
0
0 comments X

The pith

Joint adversarial policy learning with a structural penalty lets detectors handle variable outdoor illumination for garlic seedlings, raising AP50 to 91.6 percent.

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

The paper builds a new ground-based dataset of garlic seedlings captured under real field lighting that varies sharply across space and time. It introduces a framework that trains an augmentation-policy agent and an object detector together, adding a structural penalty so the learned augmentations stay realistic while still challenging the detector. The result is higher detection accuracy on the new data and better performance on the downstream task of locating missing seedlings, all without extra cost at inference time.

Core claim

Jointly optimizing a stochastic augmentation policy agent together with the object detector, subject to a structural penalty that discourages unrealistic distortions, produces illumination-robust representations that improve AP50 from the baseline to 91.6 percent on the outdoor garlic dataset and raise missing-seedling localization precision to 75.0 percent.

What carries the argument

The joint optimization of a stochastic augmentation policy agent and the object detector under a structural penalty.

If this is right

  • Detection reaches 91.6 percent AP50, 0.9 points above the plain baseline and 0.2 points above the prior best method.
  • Missing-seedling localization improves to 75.0 percent precision and 67.0 percent F1-score.
  • No extra modules are needed at inference time, keeping the detector's speed unchanged.
  • The same training procedure works for ground-based monitoring platforms without relying on greenhouse or UAV data.

Where Pith is reading between the lines

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

  • The same policy-learning loop could be applied to detection of other early-stage crops that must be monitored outdoors.
  • Replacing hand-designed illumination corrections with learned policies may reduce the need for separate preprocessing stages in agricultural vision pipelines.
  • The structural penalty offers a concrete way to keep synthetic training images inside the manifold of real field photographs.

Load-bearing premise

The learned augmentations improve robustness to real outdoor lighting changes without creating distortions that would hurt generalization on new field images.

What would settle it

Running the trained detector on a fresh collection of ground-based garlic images taken under a different set of illumination conditions and finding that the reported gains over the baseline disappear.

Figures

Figures reproduced from arXiv: 2606.26828 by Byeongkeun Kang, Chanho Kim, Soeun Lee, Yeji Kang, YoungKi Hong.

Figure 1
Figure 1. Figure 1: Examples of collected images and their annotations under diverse outdoor illumination conditions. The images include over-exposed regions, under [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed framework. During training, the policy [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the proposed training framework consisting of policy training and detection training. In the policy training stage (top), the policy network is [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of seedling detection and missing seedling localization. (a) Ground truth, (b) results of DAI-Net (Du et al., 2024), (c) results of [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
read the original abstract

Accurate seedling detection during early growth stages is essential for timely replanting and effective crop management in precision agriculture. However, existing studies are mostly evaluated under relatively stable imaging conditions, such as UAV imagery or greenhouse environments, leaving robust detection under severe and spatially heterogeneous illumination in ground-based outdoor monitoring insufficiently explored. In addition, many illumination-robust detection methods rely on additional enhancement or feature-extraction modules, which increase inference-time overhead and are not tailored to seedling detection and downstream missing seedling localization. To address these gaps, we construct a new garlic seedling dataset captured using a ground-based monitoring platform under real outdoor field conditions with highly variable illumination. We further propose an illumination-robust seedling detection framework based on adversarial augmentation policy learning. The proposed method jointly optimizes a stochastic augmentation policy agent and an object detector, enabling the detector to learn robust representations under challenging visual conditions. A structural penalty is introduced to prevent unrealistic distortions while encouraging challenging augmentations during training. Extensive experiments show that the proposed approach achieves an AP$_{50}$ of 91.6%, improving the baseline by 0.9 percentage points and outperforming the previous best-performing method by 0.2 percentage points. For downstream missing seedling localization, it achieves 75.0% precision and a 67.0% F1-score, improving the baseline by 4.8 and 2.0 percentage points, respectively. These results demonstrate the effectiveness of the proposed framework for practical ground-based agricultural monitoring under complex outdoor lighting conditions without additional inference-time computational overhead.

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 introduces a new garlic seedling dataset captured under real outdoor field conditions with highly variable illumination using a ground-based platform. It proposes an illumination-robust detection framework that jointly optimizes a stochastic augmentation policy agent with an object detector, incorporating a structural penalty to discourage unrealistic distortions while promoting challenging augmentations. Experiments report an AP50 of 91.6% (0.9 pp above baseline, 0.2 pp above prior best) and downstream missing-seedling localization gains to 75.0% precision and 67.0% F1 (improvements of 4.8 pp and 2.0 pp).

Significance. If the reported gains are shown to be robust, the work offers a practical contribution to precision agriculture by enabling reliable ground-based seedling detection without added inference-time cost. The construction of a new outdoor dataset and the adversarial policy-learning approach with an explicit structural regularizer are positive elements; however, the modest absolute improvements make rigorous validation of the central claims essential.

major comments (2)
  1. [Experimental Results] Experimental Results section: the central claim of effectiveness rests on the reported 0.9 pp AP50 gain and downstream localization improvements, yet the manuscript supplies no information on baseline methods, data splits, number of runs, standard deviations, or statistical tests; without these the numerical results cannot be verified as load-bearing evidence.
  2. [Proposed Framework] Proposed Framework section: the structural penalty is introduced to prevent unrealistic distortions, but no ablation isolating its contribution or qualitative analysis of the learned augmentations is described, leaving the weakest assumption (generalization to real illumination without distortion) untested.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'extensive experiments' is used without quantifying the scale of the evaluation (e.g., number of images, runs, or cross-validation folds).
  2. The paper does not indicate whether code or the new dataset will be released, which would strengthen reproducibility claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Experimental Results] Experimental Results section: the central claim of effectiveness rests on the reported 0.9 pp AP50 gain and downstream localization improvements, yet the manuscript supplies no information on baseline methods, data splits, number of runs, standard deviations, or statistical tests; without these the numerical results cannot be verified as load-bearing evidence.

    Authors: We agree that these details are essential for rigorous verification. We will revise the Experimental Results section to explicitly describe the baseline methods, data split protocol, number of runs, standard deviations, and statistical significance tests. revision: yes

  2. Referee: [Proposed Framework] Proposed Framework section: the structural penalty is introduced to prevent unrealistic distortions, but no ablation isolating its contribution or qualitative analysis of the learned augmentations is described, leaving the weakest assumption (generalization to real illumination without distortion) untested.

    Authors: We agree that an ablation isolating the structural penalty and qualitative analysis of the augmentations would strengthen the validation. We will add both an ablation study and qualitative examples of the learned policies in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper is an empirical ML study that constructs a new garlic seedling dataset under outdoor conditions and reports experimental performance metrics (AP50 of 91.6%, downstream precision/F1 improvements) from joint optimization of an augmentation policy and detector with a structural penalty. No equations, derivations, or self-referential predictions are present that reduce the reported results to inputs by construction, fitted parameters renamed as predictions, or load-bearing self-citations. The claims rest on standard experimental evaluation rather than any closed mathematical chain.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the effectiveness of learned adversarial augmentations plus a structural penalty for this specific detection task; no explicit free parameters, axioms, or invented entities are stated in the abstract.

pith-pipeline@v0.9.1-grok · 5816 in / 1126 out tokens · 30590 ms · 2026-06-26T05:46:18.176132+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

58 extracted references · 23 canonical work pages

  1. [1]

    LCDet: Low-Complexity Fully-Convolutional Neural Networks for Object Detection in Embedded Systems , year=

    Tripathi, Subarna and Dane, Gokce and Kang, Byeongkeun and Bhaskaran, Vasudev and Nguyen, Truong , booktitle=. LCDet: Low-Complexity Fully-Convolutional Neural Networks for Object Detection in Embedded Systems , year=

  2. [2]

    Improving weakly-supervised object localization using adversarial erasing and pseudo label , journal =

    Byeongkeun Kang and Sinhae Cha and Yeejin Lee , keywords =. Improving weakly-supervised object localization using adversarial erasing and pseudo label , journal =. 2024 , issn =. doi:https://doi.org/10.1016/j.engappai.2024.108456 , url =

    work page doi:10.1016/j.engappai.2024.108456 2024

  3. [3]

    Anandakrishnan, Jayakrishnan and Sangaiah, Arun Kumar and Darmawan, Hendri and Son, Nguyen Khanh and Lin, Yi-Bing and Alenazi, Mohammed J. F. , journal=. Precise Spatial Prediction of Rice Seedlings From Large-Scale Airborne Remote Sensing Data Using Optimized Li-YOLOv9 , year=

  4. [4]

    CoRR , volume =

    John Schulman and Filip Wolski and Prafulla Dhariwal and Alec Radford and Oleg Klimov , title =. CoRR , volume =. 2017 , url =. 1707.06347 , timestamp =

    Pith/arXiv arXiv 2017

  5. [5]

    Performance evaluation of low-power and lightweight object detectors for real-time monitoring in resource-constrained drone systems , journal =

    Murat Bakirci , keywords =. Performance evaluation of low-power and lightweight object detectors for real-time monitoring in resource-constrained drone systems , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.engappai.2025.111775 , url =

    work page doi:10.1016/j.engappai.2025.111775 2025

  6. [6]

    Joint depth-segmentation learning with segment priors for non-contact seedling height and stem thickness estimation , journal =

    Lei Song and Bo Jiang and Huaibo Song , keywords =. Joint depth-segmentation learning with segment priors for non-contact seedling height and stem thickness estimation , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.engappai.2025.111572 , url =

    work page doi:10.1016/j.engappai.2025.111572 2025

  7. [7]

    CNCAN: Contrast and normal channel attention network for super-resolution image reconstruction of crops and weeds , journal =

    Sung Jae Lee and Chaeyeong Yun and Su Jin Im and Kang Ryoung Park , keywords =. CNCAN: Contrast and normal channel attention network for super-resolution image reconstruction of crops and weeds , journal =. 2024 , issn =. doi:https://doi.org/10.1016/j.engappai.2024.109487 , url =

    work page doi:10.1016/j.engappai.2024.109487 2024

  8. [8]

    A classification method for winter wheat growth stages based on an improved version 8 of the you only look once , journal =

    Nannan Liu and Shengquan Liu and Kan Feng and Liruizhi Jia and Bo Kong , keywords =. A classification method for winter wheat growth stages based on an improved version 8 of the you only look once , journal =. 2026 , issn =. doi:https://doi.org/10.1016/j.engappai.2025.113091 , url =

    work page doi:10.1016/j.engappai.2025.113091 2026

  9. [9]

    Transformer architecture with illumination aware mechanisms for low-light image enhancement via Retinex decomposition , journal =

    Zixuan Wang and Gang Liu and Hanlin Xu and Yao Qian and Rui Chang and Durga Prasad Bavirisetti , keywords =. Transformer architecture with illumination aware mechanisms for low-light image enhancement via Retinex decomposition , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.engappai.2025.112414 , url =

    work page doi:10.1016/j.engappai.2025.112414 2025

  10. [10]

    Unsupervised detail and color restorer for Retinex-based low-light image enhancement , journal =

    Yue Sun and Yutao Jin and Xiaoyan Chen and Yanbin Xu and Xiaoning Yan and Zefu Liu , keywords =. Unsupervised detail and color restorer for Retinex-based low-light image enhancement , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.engappai.2025.110867 , url =

    work page doi:10.1016/j.engappai.2025.110867 2025

  11. [11]

    Enhancing detection performance for robotic harvesting systems through RandAugment , journal =

    Giwan Lee and Phayuth Yonrith and Doyeob Yeo and Ayoung Hong , keywords =. Enhancing detection performance for robotic harvesting systems through RandAugment , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.engappai.2023.106445 , url =

    work page doi:10.1016/j.engappai.2023.106445 2023

  12. [12]

    Image-to-Image Translation-Based Data Augmentation for Robust EV Charging Inlet Detection , year=

    Bang, Yeonjun and Lee, Yeejin and Kang, Byeongkeun , journal=. Image-to-Image Translation-Based Data Augmentation for Robust EV Charging Inlet Detection , year=

  13. [13]

    Proceedings of the 40th International Conference on Machine Learning , pages =

    Learning Instance-Specific Augmentations by Capturing Local Invariances , author =. Proceedings of the 40th International Conference on Machine Learning , pages =. 2023 , editor =

    2023

  14. [14]

    2020 , version =

    Ultralytics YOLOv5 , author =. 2020 , version =. doi:10.5281/zenodo.3908559 , orcid =

    work page doi:10.5281/zenodo.3908559 2020

  15. [15]

    International Conference on Learning Representations , year=

    AdaAug: Learning Class- and Instance-adaptive Data Augmentation Policies , author=. International Conference on Learning Representations , year=

  16. [16]

    DB-GAN: Boosting Object Recognition Under Strong Lighting Conditions , year=

    Minciullo, Luca and Manhardt, Fabian and Yoshikawa, Kei and Meier, Sven and Tombari, Federico and Kobori, Norimasa , booktitle=. DB-GAN: Boosting Object Recognition Under Strong Lighting Conditions , year=

  17. [17]

    DETRs Beat YOLOs on Real-time Object Detection , year=

    Zhao, Yian and Lv, Wenyu and Xu, Shangliang and Wei, Jinman and Wang, Guanzhong and Dang, Qingqing and Liu, Yi and Chen, Jie , booktitle=. DETRs Beat YOLOs on Real-time Object Detection , year=

  18. [18]

    2025 , eprint=

    YOLOv12: Attention-Centric Real-Time Object Detectors , author=. 2025 , eprint=

    2025

  19. [19]

    Lawrence

    Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll \'a r, Piotr and Zitnick, C. Lawrence. Microsoft COCO: Common Objects in Context. Computer Vision -- ECCV 2014. 2014

    2014

  20. [20]

    An assimilation method for wheat failure detection at the seedling stage , journal =

    Pengfei Chen and Xiao Ma and Guijun Yang , keywords =. An assimilation method for wheat failure detection at the seedling stage , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.eja.2022.126640 , url =

    work page doi:10.1016/j.eja.2022.126640 2022

  21. [21]

    Agronomy , VOLUME =

    Gao, Junpeng and Li, Yuhua and Zhou, Kai and Wu, Yanqiang and Hou, Jialin , TITLE =. Agronomy , VOLUME =. 2022 , NUMBER =

    2022

  22. [22]

    Plant Methods , VOLUME =

    Di Gennaro, Salvatore Filippo , TITLE =. Plant Methods , VOLUME =. 2020 , DOI =

    2020

  23. [23]

    Frontiers in Plant Science , VOLUME=

    Du, Xinwu and Si, Laiqiang and Jin, Xin and Li, Pengfei and Yun, Zhihao and Gao, Kaihang , TITLE=. Frontiers in Plant Science , VOLUME=. 2022 , URL=. doi:10.3389/fpls.2022.967706 , ISSN=

    work page doi:10.3389/fpls.2022.967706 2022

  24. [24]

    A novel method for detecting missing seedlings based on UAV images and rice transplanter operation information , journal =

    Shuanglong Wu and Xingang Ma and Yuxuan Jin and Junda Yang and Wenhao Zhang and Hongming Zhang and Hailin Wang and Ying Chen and Caixia Lin and Long Qi , keywords =. A novel method for detecting missing seedlings based on UAV images and rice transplanter operation information , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.compag.2024.109789 , url =

    work page doi:10.1016/j.compag.2024.109789 2025

  25. [25]

    Real-time missing seedling counting in paddy fields based on lightweight network and tracking-by-detection algorithm , journal =

    Jinrong Cui and Hong Zheng and Zhiwei Zeng and Yuling Yang and Ruijun Ma and Yuyuan Tian and Jianwei Tan and Xiao Feng and Long Qi , keywords =. Real-time missing seedling counting in paddy fields based on lightweight network and tracking-by-detection algorithm , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.compag.2023.108045 , url =

    work page doi:10.1016/j.compag.2023.108045 2023

  26. [26]

    Rapidly count crop seedling emergence based on waveform Method(WM) using drone imagery at the early stage , journal =

    Jie Yuan and Xu Li and Meng Zhou and Hengbiao Zheng and Zhitao Liu and Yang Liu and Ming Wen and Tao Cheng and Weixing Cao and Yan Zhu and Xia Yao , keywords =. Rapidly count crop seedling emergence based on waveform Method(WM) using drone imagery at the early stage , journal =. 2024 , issn =. doi:https://doi.org/10.1016/j.compag.2024.108867 , url =

    work page doi:10.1016/j.compag.2024.108867 2024

  27. [27]

    A high-precision detection method of hydroponic lettuce seedlings status based on improved Faster RCNN , journal =

    Zhenbo Li and Ye Li and Yongbo Yang and Ruohao Guo and Jinqi Yang and Jun Yue and Yizhe Wang , keywords =. A high-precision detection method of hydroponic lettuce seedlings status based on improved Faster RCNN , journal =. 2021 , issn =. doi:https://doi.org/10.1016/j.compag.2021.106054 , url =

    work page doi:10.1016/j.compag.2021.106054 2021

  28. [28]

    Machine vision-based tomato plug tray missed seeding detection and empty cell replanting , journal =

    Zeyu Yan and Yiming Zhao and Weisong Luo and Xinting Ding and Kai Li and Zhi He and Yinggang Shi and Yongjie Cui , keywords =. Machine vision-based tomato plug tray missed seeding detection and empty cell replanting , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.compag.2023.107800 , url =

    work page doi:10.1016/j.compag.2023.107800 2023

  29. [29]

    A navigation method for paddy field management based on seedlings coordinate information , journal =

    Shuanglong Wu and Zhaoguo Chen and Kemoh Bangura and Jun Jiang and Xingang Ma and Jiyu Li and Bin Peng and Xiangbao Meng and Long Qi , keywords =. A navigation method for paddy field management based on seedlings coordinate information , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.compag.2023.108436 , url =

    work page doi:10.1016/j.compag.2023.108436 2023

  30. [30]

    ByteTrack: Multi-object Tracking by Associating Every Detection Box

    Zhang, Yifu and Sun, Peize and Jiang, Yi and Yu, Dongdong and Weng, Fucheng and Yuan, Zehuan and Luo, Ping and Liu, Wenyu and Wang, Xinggang. ByteTrack: Multi-object Tracking by Associating Every Detection Box. Computer Vision -- ECCV 2022. 2022

    2022

  31. [31]

    Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks , year=

    Ren, Shaoqing and He, Kaiming and Girshick, Ross and Sun, Jian , journal=. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks , year=

  32. [32]

    Multitask AET with Orthogonal Tangent Regularity for Dark Object Detection , year=

    Cui, Ziteng and Qi, Guo-Jun and Gu, Lin and You, Shaodi and Zhang, Zenghui and Harada, Tatsuya , booktitle=. Multitask AET with Orthogonal Tangent Regularity for Dark Object Detection , year=

  33. [33]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Image-Adaptive YOLO for Object Detection in Adverse Weather Conditions , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2022 , month=. doi:10.1609/aaai.v36i2.20072 , number=

    work page doi:10.1609/aaai.v36i2.20072 2022

  34. [34]

    GDIP: Gated Differentiable Image Processing for Object Detection in Adverse Conditions , year=

    Kalwar, Sanket and Patel, Dhruv and Aanegola, Aakash and Konda, Krishna Reddy and Garg, Sourav and Krishna, K Madhava , booktitle=. GDIP: Gated Differentiable Image Processing for Object Detection in Adverse Conditions , year=

  35. [35]

    Detection-Driven Exposure-Correction Network for Nighttime Drone-View Object Detection , year=

    Xi, Yue and Jia, Wenjing and Miao, Qiguang and Feng, Junmei and Ren, Jinchang and Luo, Heng , journal=. Detection-Driven Exposure-Correction Network for Nighttime Drone-View Object Detection , year=

  36. [36]

    FeatEnHancer: Enhancing Hierarchical Features for Object Detection and Beyond Under Low-Light Vision , year=

    Hashmi, Khurram Azeem and Kallempudi, Goutham and Stricker, Didier and Afzal, Muhammamd Zeshan , booktitle=. FeatEnHancer: Enhancing Hierarchical Features for Object Detection and Beyond Under Low-Light Vision , year=

  37. [37]

    You Only Look Around: Learning Illumination-Invariant Feature for Low-light Object Detection , url =

    Hong, Mingbo and Cheng, Shen and Huang, Haibin and Fan, Haoqiang and Liu, Shuaicheng , booktitle =. You Only Look Around: Learning Illumination-Invariant Feature for Low-light Object Detection , url =. doi:10.52202/079017-2765 , editor =

    work page doi:10.52202/079017-2765

  38. [38]

    Boosting Object Detection with Zero-Shot Day-Night Domain Adaptation , year=

    Du, Zhipeng and Shit, Miaojing and Deng, Jiankang , booktitle=. Boosting Object Detection with Zero-Shot Day-Night Domain Adaptation , year=

  39. [39]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Trash to Treasure: Low-Light Object Detection via Decomposition-and-Aggregation , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2024 , month=. doi:10.1609/aaai.v38i2.27906 , number=

    work page doi:10.1609/aaai.v38i2.27906 2024

  40. [40]

    Zero-Reference Deep Curve Estimation for Low-Light Image Enhancement , year=

    Guo, Chunle and Li, Chongyi and Guo, Jichang and Loy, Chen Change and Hou, Junhui and Kwong, Sam and Cong, Runmin , booktitle=. Zero-Reference Deep Curve Estimation for Low-Light Image Enhancement , year=

  41. [41]

    Retinex-inspired Unrolling with Cooperative Prior Architecture Search for Low-light Image Enhancement , year=

    Liu, Risheng and Ma, Long and Zhang, Jiaao and Fan, Xin and Luo, Zhongxuan , booktitle=. Retinex-inspired Unrolling with Cooperative Prior Architecture Search for Low-light Image Enhancement , year=

  42. [42]

    Toward Fast, Flexible, and Robust Low-Light Image Enhancement , year=

    Ma, Long and Ma, Tengyu and Liu, Risheng and Fan, Xin and Luo, Zhongxuan , booktitle=. Toward Fast, Flexible, and Robust Low-Light Image Enhancement , year=

  43. [43]

    British Machine Vision Conference , year=

    Deep Retinex Decomposition for Low-Light Enhancement , author=. British Machine Vision Conference , year=

  44. [44]

    , title =

    Shafer, Steven A. , title =. Color , pages =. 1992 , isbn =

    1992

  45. [45]

    and Zoph, Barret and Mané, Dandelion and Vasudevan, Vijay and Le, Quoc V

    Cubuk, Ekin D. and Zoph, Barret and Mané, Dandelion and Vasudevan, Vijay and Le, Quoc V. , booktitle=. AutoAugment: Learning Augmentation Strategies From Data , year=

  46. [46]

    Online Hyper-Parameter Learning for Auto-Augmentation Strategy , year=

    Lin, Chen and Guo, Minghao and Li, Chuming and Yuan, Xin and Wu, Wei and Yan, Junjie and Lin, Dahua and Ouyang, Wanli , booktitle=. Online Hyper-Parameter Learning for Auto-Augmentation Strategy , year=

  47. [47]

    and Ghiasi, Golnaz and Lin, Tsung-Yi and Shlens, Jonathon and Le, Quoc V

    Zoph, Barret and Cubuk, Ekin D. and Ghiasi, Golnaz and Lin, Tsung-Yi and Shlens, Jonathon and Le, Quoc V. Learning Data Augmentation Strategies for Object Detection. Computer Vision -- ECCV 2020. 2020

    2020

  48. [48]

    International Journal of Computer Vision , year=

    Li, Lin and Qiu, Jianing and Spratling, Michael , title=. International Journal of Computer Vision , year=. doi:10.1007/s11263-024-02206-4 , url=

    work page doi:10.1007/s11263-024-02206-4

  49. [49]

    International Conference on Learning Representations , year=

    Adversarial AutoAugment , author=. International Conference on Learning Representations , year=

  50. [50]

    1977 , month=

    The Retinex Theory of Color Vision , journal=. 1977 , month=. doi:10.1038/scientificamerican1277-108 , author=

    work page doi:10.1038/scientificamerican1277-108 1977

  51. [51]

    Williams

    Williams, Ronald J. , title=. Machine Learning , year=. doi:10.1007/BF00992696 , url=

    work page doi:10.1007/bf00992696

  52. [52]

    Identity Mappings in Deep Residual Networks

    He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian. Identity Mappings in Deep Residual Networks. Computer Vision -- ECCV 2016. 2016

    2016

  53. [53]

    and Kim, Seungryong and Choo, Jaegul , booktitle=

    Choi, Sungha and Jung, Sanghun and Yun, Huiwon and Kim, Joanne T. and Kim, Seungryong and Choo, Jaegul , booktitle=. RobustNet: Improving Domain Generalization in Urban-Scene Segmentation via Instance Selective Whitening , year=

  54. [54]

    Lempitsky , title =

    Dmitry Ulyanov and Andrea Vedaldi and Victor S. Lempitsky , title =. CoRR , volume =. 2016 , url =. 1607.08022 , timestamp =

    Pith/arXiv arXiv 2016

  55. [55]

    and Sheikh, H.R

    Zhou Wang and Bovik, A.C. and Sheikh, H.R. and Simoncelli, E.P. , journal=. Image quality assessment: from error visibility to structural similarity , year=

  56. [56]

    and Huang, Weilin , booktitle=

    Feng, Chengjian and Zhong, Yujie and Gao, Yu and Scott, Matthew R. and Huang, Weilin , booktitle=. TOOD: Task-aligned One-stage Object Detection , year=

  57. [57]

    CoRR , volume =

    Joseph Redmon and Ali Farhadi , title =. CoRR , volume =. 2018 , url =. 1804.02767 , timestamp =

    Pith/arXiv arXiv 2018

  58. [58]

    Correcting over-exposure in photographs , year=

    Guo, Dong and Cheng, Yuan and Zhuo, Shaojie and Sim, Terence , booktitle=. Correcting over-exposure in photographs , year=