Pith. sign in

REVIEW 5 major objections 5 minor 35 references

Adaptive Object Detection with ESRGAN-Enhanced Resolution & Faster R-CNN

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Upscaling low-res images with ESRGAN before Faster R-CNN detection lifts accuracy to 89 percent.

desk verdict The paper's only real claim, the 89% accuracy from ESRGAN+Faster R-CNN, is unsupported by any experimental protocol and is internally contradicted by the paper's own figures. read the letter →

arxiv 2506.11122 v1 pith:44AR34VA submitted 2025-06-10 cs.CV

Divya Swetha K , Ziaul Haque Choudhury , Hemanta Kumar Bhuyan , Biswajit Brahma , Nilayam Kumar Kamila This is my paper
classification cs.CV
keywords objectdetectionlow-resolutionimagesESRGANFasterR-CNNsuper-resolutionimageenhancementbenchmarkevaluation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a two-stage pipeline—upscale a low-resolution image with ESRGAN, then detect objects with Faster R-CNN—beats running either model alone or applying a traditional detector to the raw low-resolution input. The reported numbers are 89% accuracy, 87% precision, and 85% recall for the combined pipeline, compared with 78%, 75%, and 73% for Faster R-CNN alone and 65%, 60%, and 55% for the traditional baseline. If the claim holds, it matters because degraded inputs are common in surveillance, medical imaging, and satellite analysis, and the result suggests that improving image quality before detection can substitute for more complex detector changes. The paper presents enhancement-before-detection as a general recipe rather than a task-specific fix.

What carries the argument

The mechanism is the serial composition of two networks. ESRGAN's generator, built from Residual-in-Residual Dense Blocks with a skip connection and upsampling layer, learns a mapping $G$ from a low-resolution image $I_{LR}$ to a high-resolution image $I_{HR}=G(I_{LR})$, trained with adversarial loss and perceptual loss so the output is realistic as well as pixel-accurate. Faster R-CNN then takes $I_{HR}$ through a Region Proposal Network that produces class probabilities and box refinements, and RoI pooling that normalizes proposals for final classification and bounding-box regression. The load-bearing identity is the composition of the detector applied to $G(I_{LR})$: the detector never sees the original low-resolution image, only the super-resolved one.

What would settle it

Train Faster R-CNN on original low-resolution images and evaluate it on the same low-resolution test set, then compare against Faster R-CNN applied to ESRGAN-upscaled versions; if the native-resolution model reaches or exceeds the reported 89% accuracy, 87% precision, and 85% recall, the claim that enhancement before detection causes the gain is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that the composition of ESRGAN and Faster R-CNN, applied as a fixed sequence, produces better object detection on low-resolution images than either component alone or any single detector in its comparison table. Concretely, the authors report 89% accuracy, 87% precision, and 85% recall for the combined pipeline, a gain of 11 accuracy points over Faster R-CNN alone and 24 points over the traditional baseline. The authors interpret this as showing that image enhancement before detection leads to better results, and they ground the claim in the 80-category benchmark dataset used for the experiments.

Load-bearing premise

The pipeline assumes that the details ESRGAN fabricates during upscaling preserve or improve the object-discriminative information Faster R-CNN relies on, rather than introducing artifacts that degrade detection.

Editorial extensions

If this is right

  • On the reported metrics, the combined pipeline outperforms Faster R-CNN alone by 11 accuracy points and 12 points in precision and recall, and it also exceeds every single detector listed in the paper's table, including the strongest at 87%.
  • Because ESRGAN sits before the detector as a preprocessing step, the pipeline can be deployed on any low-resolution input without modifying the detection network itself.
  • The reported gain indicates that image-restoration quality is a first-order lever for detection performance on degraded inputs, not just a cosmetic preprocessing step.
  • The improvement reported for the two-stage pipeline supports applying the same sequence to surveillance, wildlife monitoring, and other settings where input resolution is variable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: the decisive control experiment would train Faster R-CNN on native low-resolution images and compare it with the ESRGAN-upscaled path on the same test set, isolating the contribution of enhancement from the contribution of detector capacity.
  • Beyond the paper: the same enhancement-then-detect composition could be tested with other detectors to see whether the benefit is specific to Faster R-CNN or generalizes to anchor-free and transformer detectors.
  • Beyond the paper: reporting accuracy as a function of input resolution and object size would show whether ESRGAN's benefit is concentrated on small objects, where fabricated detail is most likely to help or hurt.
  • Beyond the paper: a public reproduction with a fixed resolution-degradation protocol, and with training setups stated for every compared model, would let the reported figures transfer to other applications.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a two-stage object-detection pipeline in which a low-resolution input is first processed by ESRGAN super-resolution and then passed to a Faster R-CNN detector. The authors report a peak performance of 89% accuracy, 87% precision, and 85% recall for the combined system, and they interpret this as evidence that image enhancement before detection is beneficial. The methods section presents textbook formulations of ESRGAN's adversarial, perceptual, and total losses and a standard description of Faster R-CNN. The experimental section, however, provides no measurable protocol: there is no dataset split, no low-resolution degradation model, no training or inference configuration, no definition of the baseline 'Traditional Model', and the figures numerically contradict the tables.

Significance. If the reported empirical claim were properly supported, the paper would document a straightforward and potentially useful engineering result: that a generic super-resolution front-end can improve a standard detector on low-resolution inputs. However, the manuscript does not supply the experimental evidence needed to validate this claim. There are no reproducible artifacts, no code, no data, no hyperparameters, and no statistical uncertainty. The internal inconsistencies between Table II and Figures 6 and 7 further undermine the single load-bearing result. As written, the paper contributes only a restatement of well-known components and an unverifiable performance table; the significance is therefore limited to the plausibility of the idea, not to any demonstrated outcome.

major comments (5)
  1. [Section IV, Tables I and II] The central claim of 89% accuracy, 87% precision, and 85% recall is reported without any experimental protocol. Please specify the dataset split (training/validation/test), the method used to generate low-resolution inputs (including the downscaling factor and interpolation kernel), the Faster R-CNN backbone and training schedule, the ESRGAN inference configuration, and the exact definition of the reported metrics. Without these details, the numbers in Table II cannot be reproduced or checked.
  2. [Section IV, paragraph after Table II] The statement that the result 'proving that image enhancement before detection leads to better results' is not supported by the experimental design, because there is no control condition consisting of Faster R-CNN trained and evaluated directly on native low-resolution images. The comparison in Table II rows 3 and 4 confounds the effect of ESRGAN preprocessing with potential differences in training or evaluation setup, so the causal conclusion is unjustified.
  3. [Section IV, Figures 6 and 7] Figures 6 and 7 numerically contradict Table II. Figure 6 reports relative shares of 21%, 25%, 25%, and 29% for the four methods, which do not match the accuracy values 65%, 75%, 78%, and 89% in Table II. Figure 7 uses an axis ranging to 350 with categories 'Increase/Decrease/Total' that are incompatible with accuracy expressed as a percentage. These inconsistencies make the reported results impossible to verify and must be corrected with a single consistent set of numbers.
  4. [Section III.A, Equations (2)–(4)] The text states that ESRGAN's total loss incorporates content loss, perceptual loss, and adversarial loss, but Equation (4) defines the total loss as only a weighted sum of the GAN loss and the perceptual loss, with no content-loss term. The content loss is also never defined. This is a technical gap in the description of the core method and should be corrected by either adding the missing term and definition or revising the surrounding text.
  5. [Section IV, Table I] The 'Traditional Model' baseline is never defined. Without knowing what this baseline is (architecture, training data, input resolution), the comparative claims of improvement over a 'traditional' approach are vacuous. Please provide a concrete description of the baseline model and its configuration.
minor comments (5)
  1. [Abstract] The first sentence lacks a grammatical subject ('In this study, proposes a method') and the phrase 'image resolution is in consistent' should be 'image resolution is inconsistent'.
  2. [Section III.B] The terms 'ROI Pooling' and 'RoI Pooling' are used interchangeably; please use a consistent notation.
  3. [Figure 5 caption] The caption states that the x-axis enumerates object detection architectures, but the body text and the figure itself appear to show accuracy, precision, and recall bars; please align the caption, axis labels, and the values being compared.
  4. [References] Reference [35] is malformed: the author list is incomplete and the URL is truncated ('...htmlhttps://openaccess.thecvf.com/...'). Please provide the complete bibliographic entry.
  5. [Throughout] The manuscript does not state whether any code, trained models, or computed evaluation logs will be made available, which would be essential for reproducibility given the unverified experimental numbers.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper presents a two-stage pipeline with standard loss equations and asserted performance numbers, with no derivation chain that reduces a prediction to its own inputs.

full rationale

The manuscript contains no derivation chain of the kind that could be circular. The loss equations in Section III (adversarial loss, perceptual loss, and total loss) are standard textbook definitions taken from ESRGAN and Faster R-CNN literature, and the reported 89% accuracy in Tables I and II is asserted as an experimental result rather than derived from a fitted parameter or from the equations themselves. There is no evidence of a parameter fitted to a subset of data and then renamed as a prediction, no uniqueness theorem imported from the authors' prior work, and no ansatz smuggled in via self-citation. The paper's central claim that ESRGAN preprocessing improves Faster R-CNN detection is empirically unsupported as written, because no experimental protocol, dataset split, degradation model, or native low-resolution control condition is reported, and the figures are internally inconsistent with the tables. However, unsupported or unverifiable empirical claims are a correctness and reproducibility problem, not a circularity problem. The circularity burden requires showing that the conclusion is equivalent to an input by construction, and no such reduction is present or quotable in this manuscript.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central claim rests on standard off-the-shelf components and an implicit, undisclosed evaluation protocol. No new entity is introduced. The only explicit hand-set quantities are the loss weights in Eq. (4), which are never assigned values.

free parameters (2)
  • lambda_GAN = not reported
    Weight on adversarial loss in Eq. (4); its value affects the super-resolution output that feeds detection, but no value or tuning procedure is given.
  • lambda_perceptual = not reported
    Weight on perceptual loss in Eq. (4); same issue as lambda_GAN, with no reported value or selection method.
assumptions (3)
  • domain assumption ESRGAN and Faster R-CNN are used as off-the-shelf models exactly as described in their original publications.
    Section III describes both components only at a high level; the reported performance depends on pretrained weights, architecture versions, and training recipes that are not specified.
  • domain assumption An implicit MS COCO training and evaluation protocol underlies Tables I and II, comparable across all listed methods.
    Section IV describes MS COCO but never states the split, low-resolution degradation model, or evaluation metric conventions used to produce the accuracy, precision, and recall numbers.
  • domain assumption Super-resolving low-resolution images before detection preserves or improves all task-relevant information.
    The central improvement claim relies on this premise, which is not tested against the alternative of training a detector directly on low-resolution inputs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Object Detection with ESRGAN-Enhanced Resolution & Faster R-CNN." pith.science (2026). https://pith.science/paper/44AR34VA

@misc{pith2026250611122,
  author       = {Pith},
  title        = {Pith review of: Adaptive Object Detection with ESRGAN-Enhanced Resolution & Faster R-CNN},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/44AR34VA}},
  note         = {Machine review of arXiv:2506.11122}
}
read the original abstract

In this study, proposes a method for improved object detection from the low-resolution images by integrating Enhanced Super-Resolution Generative Adversarial Networks (ESRGAN) and Faster Region-Convolutional Neural Network (Faster R-CNN). ESRGAN enhances low-quality images, restoring details and improving clarity, while Faster R-CNN performs accurate object detection on the enhanced images. The combination of these techniques ensures better detection performance, even with poor-quality inputs, offering an effective solution for applications where image resolution is in consistent. ESRGAN is employed as a pre-processing step to enhance the low-resolution input image, effectively restoring lost details and improving overall image quality. Subsequently, the enhanced image is fed into the Faster R-CNN model for accurate object detection and localization. Experimental results demonstrate that this integrated approach yields superior performance compared to traditional methods applied directly to low-resolution images. The proposed framework provides a promising solution for applications where image quality is variable or limited, enabling more robust and reliable object detection in challenging scenarios. It achieves a balance between improved image quality and efficient object detection

Figures

Figures reproduced from arXiv: 2506.11122 by the authors.

Figure 1
Figure 1. shows the proposed system architecture. A. Generative Adversarial Network with Enhanced Super Resolution (ESRGAN) The ESRGAN module is in charge of sharpening the features and recovering lost details in order to transform a low-resolution image into a high-resolution one. It learns significant textures and minute details from photos using Residual-in-Residual Dense Blocks (RRDBs) [35]. A skip connection helps the ne… view at source ↗
Figure 6
Figure 6. illustrates the accuracy distribution among four [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 5
Figure 5. illustrates the performance comparison of various [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figures from the paper (1 more)
Figure 8
Figure 8. Figure 8: These pictures show how a computer can learn to [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 23 canonical work pages

  1. [1]

    A review of object detection techniques,

    K. and L. Cao, "A review of object detection techniques," presented at the 2020 5th Int. Conf. Electromechanical Control Technology and Transportation (ICECTT), 2020

    work page 2020

  2. [2]

    Object detection in 20 years: A survey,

    Z. Zou, Z. Shi, Y . Guo, and J. Ye, "Object detection in 20 years: A survey," arXiv preprint arXiv:1905.05055 , 2019. [Online]. Available: https://arxiv.org/abs/1905.05055

    arXiv 1905

  3. [3]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," Commun. ACM , vol. 60, no. 6, pp. 84 –90, 2017. [Online]. Available: https://doi.org/10.1145/3065386

    doi:10.1145/3065386 2017

  4. [4]

    Recent object detection techniques: A survey,

    D. Diwakar and D. Raj, "Recent object detection techniques: A survey," Int. J. Image Graphics Signal Process. , vol. 14, no. 2, pp. 47 –60, 2022, doi: 10.5815/ijigsp.2022.02.05. [Online]. Available: http://www.mecs-press.org/ijigsp/ijigsp-v14-n2/

    work page doi:10.5815/ijigsp.2022.02.05 2022

  5. [5]

    Deep learning for generic object detection: A survey,

    L. Liu, W. Ouyang, X. Wang, P . Fieguth, J. Chen, X. Liu, and M. Pietikäinen, "Deep learning for generic object detection: A survey," Int. J. Comput. Vis., vol. 128, no. 2, pp. 261–318, 2020. [Online]. Available: https://doi.org/10.1007/s11263-019-01230-x

    work page doi:10.1007/s11263-019-01230-x 2020

  6. [6]

    Object detection techniques: Overview and performance comparison,

    M. Noman, V . Stankovic, and A. Tawfik, "Object detection techniques: Overview and performance comparison," presented at the 2019 IEEE Int. Symp. Signal Process. Inform. Technol. (ISSPIT), 2019. [Online]. Available: https://ieeexplore.ieee.org/document/8957477

    work page arXiv 2019

  7. [7]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs,

    L. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. Yuille, "Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs," IEEE Trans. Pattern Anal. Mach. Intell., vol. 40, no. 4, pp. 834–848, 2018, doi: 10.1109/tpami.2017.2699184. [Online]. Available: https://doi.org/10.1109/tpami.2017.2699184

    arXiv 2018

  8. [8]

    Image resolution enhancement by using discrete and stationary wavelet decomposition,

    H. Demirel and G. Anbarjafari, "Image resolution enhancement by using discrete and stationary wavelet decomposition," IEEE Trans. Image Process., vol. 20, no. 5, pp. 1458–1460, 2011, doi: 10.1109/tip.2010.2087767. [Online]. Available: https://doi.org/10.1109/tip.2010.2087767

    work page arXiv 2011

Show all 35 references
  1. [9]

    Detection and classification of multiple objects using an RGB - D sensor and linear spatial pyramid matching,

    M. Dimitriou, T. Kounalakis, N. Vidakis, and G. Triantafyllidis, "Detection and classification of multiple objects using an RGB - D sensor and linear spatial pyramid matching," Elcvia Electron. Lett. Comput. Vis. Image Anal., vol. 12, no. 2, pp. 78 –87, 2013, doi: 10.5565/rev/...

    2013 doi

  2. [10]

    Artificial intelligence system of faster region -based convolutional neural network surpassing senior radiologists in evaluation of metastatic lymph nodes of rectal cancer,

    L. Ding, G. Liu, B. Zhao, Y . Zhou, S. Li, Z. Zhang, and L. Wang, "Artificial intelligence system of faster region -based convolutional neural network surpassing senior radiologists in evaluation of metastatic lymph nodes of rectal cancer," China Med. J. , vol. 132, no. 4, pp....

    2019 doi

  3. [11]

    Super resolution image reconstruction using wavelet lifting schemes and Gabor filters,

    S. Dogiwal, Y . Shishodia, and A. Upadhyaya, "Super resolution image reconstruction using wavelet lifting schemes and Gabor filters," presented at the 2014 IEEE Conf. Confluence, pp. 625 – 630, 2014, doi: 10.1109/confluence.2014.6949252. [Online]. Available: https://ieeexplore...

    2014

  4. [12]

    Object detection with discriminatively trained part -based models,

    P. Felzenszwalb, R. Girshick, D. McAllester, and D. Ramanan, "Object detection with discriminatively trained part -based models," IEEE Trans. Pattern Anal. Mach. Intell., vol. 32, no. 9, pp. 1627 –1645, 2010, doi: 10.1109/tpami.2009.167. [Online]. Available: https://doi.org/10...

    2010 doi

  5. [13]

    Vehicle overtaking hazard detection over onboard cameras using deep convolutional networks,

    J. García -González, I. García -Aguilar, D. Medina, R. Luque - Baena, E. López-Rubio, and E. Domínguez, "Vehicle overtaking hazard detection over onboard cameras using deep convolutional networks," in Proc. of the 2022 Int. Conf. on Computer Vision and Image Processing, pp. 33...

    2022 doi

  6. [14]

    Fast R -CNN,

    R. Girshick, "Fast R -CNN," presented at the IEEE Int. Conf. Comput. Vis. (ICCV), 2015, doi: 10.1109/iccv.2015.169. [Online]. Available: https://doi.org/10.1109/iccv.2015.169

    2015 doi

  7. [15]

    Mask R-CNN,

    K. He, G. Gkioxari, P. Dollár, and R. Girshick, "Mask R-CNN," presented at the IEEE Int. Conf. Comput. Vis. (ICCV), 2017, doi: 10.1109/iccv.2017.322. [Online]. Available: https://doi.org/10.1109/iccv.2017.322

    2017 doi

  8. [16]

    A comprehensive survey of video datasets for background subtraction,

    R. Kalsotra and S. Arora, "A comprehensive survey of video datasets for background subtraction," IEEE Access, vol. 7, pp. 59143–59171, 2019, doi: 10.1109/access.2019.2914961. [Online]. Available: https://ieeexplore.ieee.org/document/8914961

    2019

  9. [17]

    Evaluation of the accuracy of oil palm tree detection using deep learning and support vector machine classifiers,

    N. Khalid and N. Shahrol, "Evaluation of the accuracy of oil palm tree detection using deep learning and support vector machine classifiers," IOP Conf. Ser . Earth Environ. Sci., vol. 1051, no. 1, p. 012028, 2022, doi: 10.1088/1755 -1315/1051/1/012028. [Online]. Available: htt...

    2022 doi

  10. [18]

    Accurate image super -resolution using very deep convolutional networks,

    J. Kim, J. Lee, and K. Lee, "Accurate image super -resolution using very deep convolutional networks," presented at the IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, doi: 10.1109/cvpr.2016.182. [Online]. Available: https://doi.org/10.1109/cvpr.2016.182

    2016 doi

  11. [19]

    Image resolution enhancement using convolutional autoencoders,

    P. Kumar and N. Goel, "Image resolution enhancement using convolutional autoencoders," Energies, vol. 7, no. 8, pp. 8259 – 8284, 2020, doi: 10.3390/ecsa -7-08259. [Online]. Available: https://doi.org/10.3390/ecsa-7-08259

    2020 doi

  12. [20]

    63‐3: real‐time spatial‐based projector resolution enhancement,

    A. Ma, A. Gawish, M. Lamm, A. Wong, and P. Fieguth, "63‐3: real‐time spatial‐based projector resolution enhancement," SID Symp. Digest of Technical Papers , vol. 49, no. 1, pp. 831 –834, 2018, doi: 10.1002/sdtp.12243. [Online]. Available: https://doi.org/10.1002/sdtp.12243

    2018 doi

  13. [21]

    Deep learning vs. traditional computer vision,

    N. Mahony et al., "Deep learning vs. traditional computer vision," in Springer Handbook of Computational Intelligence , 2019, pp. 128-144. [Online]. Available: https://doi.org/10.1007/978-3-030- 17795-9_10

    2019 doi

  14. [22]

    A review: object detection using deep learning,

    Z. Naik and M. Gandhi, "A review: object detection using deep learning," Int. J. Comput. Appl., vol. 180, no. 29, pp. 46–48, 2018, doi: 10.5120/ijca2018916708. [Online]. Available: https://doi.org/10.5120/ijca2018916708

    2018 doi

  15. [23]

    Frustum PointNets for 3D object detection from RGB-D data,

    C. Qi, W. Liu, C. Wu, H. Su, and L. Guibas, "Frustum PointNets for 3D object detection from RGB-D data," presented at the IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, doi: 10.1109/cvpr.2018.00102. [Online]. Available: https://doi.org/10.1109/cvpr.2018.00102

    2018

  16. [24]

    Comparative analysis of various enhancement methods for satellite images,

    M. Reddy, "Comparative analysis of various enhancement methods for satellite images," CVR J. Sci. Technol., vol. 7, no. 1, pp. 69 –72, 2014. [Online]. Available: https://doi.org/10.32377/cvrjst0712

    2014 doi

  17. [25]

    Fish detection using morphological approach based -on K -means segmentation,

    S. Saifullah, A. Suryotomo, and B. Y uwono, "Fish detection using morphological approach based -on K -means segmentation," Compiler, vol. 10, no. 1, 2021. [Online]. Available: https://doi.org/10.28989/compiler.v10i1.946

    2021 doi

  18. [26]

    PSO -based Tsallis thresholding selection procedure for image segmentation,

    P. Sathya and R. Kayalvizhi, "PSO -based Tsallis thresholding selection procedure for image segmentation," Int. J. Comput. Appl., vol. 5, no. 4, pp. 39 –46, 2010, doi: 10.5120/903 -1279. [Online]. Available: https://doi.org/10.5120/903-1279

    2010 doi

  19. [27]

    Deep learning techniques for image recognition and object detection,

    K. Sharada, "Deep learning techniques for image recognition and object detection," E3S Web Conf., vol. 399, p. 04032, 2023, doi: 10.1051/e3sconf/202339904032. [Online]. Available: https://doi.org/10.1051/e3sconf/202339904032

    2023

  20. [28]

    Colorectal polyp detection model by using super - resolution reconstruction and YOLO,

    S. Wang, "Colorectal polyp detection model by using super - resolution reconstruction and YOLO," Electronics, vol. 13, no. 12, p. 2298, 2024, doi: 10.3390/electronics13122298. [Online]. Available: https://doi.org/10.3390/electronics13122298

    2024 doi

  21. [29]

    Super-resolution of plant disease images for the acceleration of image -based phenotyping and vigor diagnosis in agriculture,

    K. Yamamoto, T. Togami, and N. Yamaguchi, "Super-resolution of plant disease images for the acceleration of image -based phenotyping and vigor diagnosis in agriculture," Sensors, vol. 17, no. 11, p. 2557, 2017, doi: 10.3390/s17112557. [Online]. Available: https://doi.org/10.33...

    2017 doi

  22. [30]

    Combined self -learning based single -image super - resolution and dual-tree complex wavelet transform denoising for medical images,

    G. Yang, X. Ye, G. Slabaugh, J. Keegan, R. Mohiaddin, and D. Firmin, "Combined self -learning based single -image super - resolution and dual-tree complex wavelet transform denoising for medical images," Proc. SPIE, vol. 9784, p. 97840L, 2016, doi: 10.1117/12.2207440. [Online]...

    2016 doi

  23. [31]

    Unsupervised vehicle extraction of bounding boxes in UA V images,

    J. Yeom, "Unsupervised vehicle extraction of bounding boxes in UA V images," presented at the SPIE Remote Sensing, 2023, doi: 10.1117/12.2680067. [Online]. Available: https://doi.org/10.1117/12.2680067

    2023 doi

  24. [32]

    3D object detection method based on YOLO and K -means for image and point clouds,

    X. Yin, Y . Sasaki, W. Wang, and K. Shimizu, "3D object detection method based on YOLO and K -means for image and point clouds," arXiv preprint arXiv:2005.02132 , 2020. [Online]. Available: https://arxiv.org/abs/2005.02132

    2005 arXiv

  25. [33]

    Lesion focused super -resolution,

    J. Zhu, G. Yang, and P. Lió, "Lesion focused super -resolution," Proc. SPIE, vol. 56, 2019, doi: 10.1117/12.2512576. [Online]. Available: https://doi.org/10.1117/12.2512576

    2019 doi

  26. [34]

    Microsoft Coco : Common objects in context,

    T.-Y . Lin et al., “Microsoft Coco : Common objects in context,” in proc, Eur. Conf. Comput. Vis. Cham, Switzerland : springer, 2014, pp.740-755

    2014

  27. [35]

    Xintao Wang1, Ke Yu1, Shixiang Wu2,”Wang_ESRGAN_Enhanced_Super- Resolution_Generative_Adversarial_Networks_ECCVW_2018_ paper”.htmlhttps://openaccess.thecvf.com/content_eccv_2018_ workshops/w25/html/

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.