Pith. sign in

REVIEW 3 major objections 5 minor 58 references

VisAlgae 2023: A Dataset and Challenge for Algae Detection in Microscopy Images

T0 review · 3 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 This paper introduces VisAlgae 2023, a public 1,000-image benchmark for detecting six microalgae genera in microscopy images, and reports that lightweight detectors outperform larger models while the winning challenge entry reaches 0.7604…

desk verdict A genuine dataset and challenge summary for a niche but real application; the numbers are less clean than they look, but the dataset itself deserves peer review with revisions. read the letter →

arxiv 2505.20687 v1 pith:YAVIL35T submitted 2025-05-27 cs.CV

Mingxuan Sun , Juntao Jiang , Zhiqiang Yang , Shenao Kong , Jiamin Qi , Jianru Shang , Shuangling Luo , Wanfa Sun
show 16 more authors
Tianyi Wang Yanqi Wang Qixuan Wang Tingjian Dai Tianxiang Chen Jinming Zhang Xuerui Zhang Yuepeng He Pengcheng Fu Qiu Guan Shizheng Zhou Yanbo Yu Qigui Jiang Teng Zhou Liuyong Shi Hong Yan
This is my paper
classification cs.CV
keywords microalgaedetectionmicroscopyimagedatasetobjectbenchmarksmalldomainshiftdataaugmentationVisAlgaechallengemAP50:95
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

The paper establishes VisAlgae 2023, a public benchmark for detecting microalgae cells in high-throughput microscopy images: 1,000 expert-annotated images covering six genera, split 700/300 into training and test sets with test annotations kept private during evaluation. Its baseline experiments show that on this domain detection accuracy does not grow with model size: the lightweight YOLOv5s reaches $0.711$ mAP$_{50:95}$ on the validation split and beats its larger siblings, while newer YOLO generations trained on natural images transfer poorly to the translucent, low-contrast appearance of microalgae. The top ten challenge entries, drawn from 369 participating teams, all relied on high-resolution inputs, strong copy-paste and Poisson-blend augmentation, multi-scale training, and weighted-box-fusion ensembling; the winning submission reached $0.7604$ mAP$_{50:95}$ on the private test set. A sympathetic reader would care because the dataset turns a manual expert task into a standardized computer-vision benchmark and gives concrete, transferable guidance for detecting small objects in microscopy.

What carries the argument

The central object is the VisAlgae 2023 benchmark: a public set of 1,000 microscopy images of six microalgae genera, collected on a high-throughput microfluidic platform and manually annotated with bounding boxes, with a 700/300 train/test split and private test annotations. The evaluation protocol is mean average precision averaged over Intersection-over-Union thresholds from $0.5$ to $0.95$ (mAP$_{50:95}$), the metric used to rank all baselines and challenge submissions. This dataset-plus-metric machinery carries the argument because it lets models of different sizes, families, and training strategies be compared on the same fine-grained, small-object detection task, which is what makes the paper's claims about scale, domain shift, and augmentation concrete.

What would settle it

Run the top-ranked models on a fresh set of annotated algae images taken with the same microscope and platform but never released during the challenge, then recompute mAP$_{50:95}$; if the ordering of teams changes sharply or the scores fall well below the reported $0.7604$, the leaderboard numbers partly measure test-set adaptation rather than generalization.

Watch

Extended reading notes

Core claim

The paper's central claim is that the VisAlgae 2023 dataset exposes a real domain gap between natural-image object detection and microscopy-based algae counting, and that within this domain lightweight detectors can match or exceed much larger models. The authors demonstrate this with a systematic baseline sweep across YOLOv5, YOLOv8, YOLOv9, YOLOv10, and YOLOv11, where neither newer architecture nor larger parameter count consistently improves mAP$_{50:95}$; the best baseline behavior appears at intermediate or small scale, and all models perform worst on Chlorella, the class with the smallest annotation boxes. They further report that the top-performing competition entries, which reached $0.7604$, achieve their gains mostly through data-centric techniques—high input resolution, Poisson fusion and copy-paste augmentation, multi-scale training, test-time augmentation, and weighted box fusion—rather than through novel detection architectures. The paper states these findings as evidence that dataset-specific fine-tuning and small-object handling matter more than blind reliance on generic pretrained models.

Load-bearing premise

The load-bearing premise is that the public leaderboard scores measure real detection skill and not repeated peeking at the test set, because teams could submit many times and overwrite their results over a month.

Editorial extensions

If this is right

  • On microscopy data, model size is not a reliable proxy for accuracy: a small YOLOv5 variant outperforms larger versions, so deployment choices should be made per-domain rather than by parameter count alone.
  • Newer detectors optimized for natural images can underperform older ones on translucent, low-contrast microscopic targets, implying that benchmark progress on MS COCO does not automatically transfer to specialized imaging domains.
  • Small-object handling—higher input resolution, P2 feature layers, and multi-scale training—is the most consistent lever for improving algae detection, especially for the tiny Chlorella class.
  • Ensembling and test-time augmentation provide the largest accuracy gains in the reported solutions, but the paper also notes these are often too slow for real-time deployment, pointing to a trade-off between leaderboard score and practical use.
  • The dataset is released under a permissive license, allowing other researchers to reuse and extend it for cross-modality adaptation and further benchmarking.

Reading between the lines

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

  • Because the evaluation allowed up to five overwriting submissions per day for more than a month, the leaderboard probably includes some degree of adaptation to the public test set; testing the same models on a newly annotated batch would separate generalization from test-set feedback.
  • The winning solution's dynamic cache of algae crops for Poisson blending suggests that treating the training set as a reusable pool of object appearances is a promising direction for other small-object microscopy tasks; a controlled study could test this recipe outside algae.
  • The paper's finding that all models struggle most on the smallest class implies the benchmark may be better described as a small-object detection problem than an algae-classification problem; reporting size-stratified mAP would sharpen future comparisons.
  • The challenge's emphasis on mAP alone, as the paper itself notes, leaves efficiency out of the ranking; a metric that includes latency would likely favor the lightweight detectors the baseline section highlights.
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

3 major / 5 minor

Summary. The paper presents the VisAlgae 2023 challenge and dataset for microalgae detection in microscopy images: 1000 images, six algae classes, a 700/300 train/test split, and mAP50:95 as the evaluation metric. The authors report baseline experiments with YOLOv5/8/9/10/11 variants and summarize the top-10 leaderboard methods, emphasizing the role of high-resolution inputs, augmentation (Poisson fusion, copy-paste, Mixup), attention modules, Weighted Box Fusion, and Test-Time Augmentation. The central claims are that the dataset is a useful public benchmark for an under-explored domain and that lightweight detectors can match or outperform larger models on this task.

Significance. If the benchmark numbers are reliable, this is a valuable public resource for microalgae detection: the dataset is released under CC-BY-4.0, it covers six taxonomically meaningful classes with small-object and multi-scale challenges, and the challenge attracted 369 teams. The structured summary of the top-10 methods is a useful practical reference for practitioners. The paper's main weaknesses are that the baseline results are not computed on the same test set as the leaderboard and that the public-leaderboard protocol permits repeated feedback-driven submissions, so the headline comparisons are not yet fully trustworthy. The dataset contribution itself is solid and could support a strong revised version.

major comments (3)
  1. [Sections 4.1 and 4.2] The baseline results are computed on an 8:2 split of the 700 training images, while the leaderboard scores in Table 3 are computed on the 300-image test set described in Section 3.1.1. The claim in Section 4.2 that 'the lightweight variant YOLOv5 achieved better performance (0.711) compared to its larger counterparts (0.69–0.70)' is therefore a validation-split comparison, not a challenge-leaderboard comparison. The paper should either evaluate the baselines on the same hidden test set or explicitly state that the comparison is validation-only and not directly tied to Table 3. Additionally, no error bars or multiple-seed runs are reported, so the small differences between models could be within run-to-run variability.
  2. [Section 3.2 and Table 3] The evaluation protocol allows each team to submit up to five times per day between December 22, 2023, and January 25, 2024, with automatic overwriting and no described total-submission cap or final private test set. Because the leaderboard exposes the score after each submission, teams can iteratively probe the test set, and the final ranking may reflect selection over many attempts rather than a single held-out evaluation. The gap between rank 1 (0.7604) and rank 10 (0.7244) is only 0.036 mAP, so even modest selection bias could reorder the leaderboard. The paper needs to address this directly, for example by reporting the number of submissions per team, comparing early versus late scores, using a final hidden re-evaluation, or providing an analysis of the stability of the scores.
  3. [Section 5, Tables 5–10] The ablation studies and incremental gains reported by the top-10 teams (e.g., the 0.46% gain from AIFI and 0.66% gain from CARAFE in Table 9, or the RepCSPLayer results in Table 8) are self-reported with no checkpoints, code, or repeated-run statistics. The paper presents these numbers as evidence for the effectiveness of the described techniques. The text should explicitly state that these are unreviewed self-reports from participants and not independently verified, or the authors should provide reproducibility materials for the participant methods.
minor comments (5)
  1. [Section 5.2.2 and Table 9] The module name 'CAREFE' appears to be a typo for 'CARAFE'; the same typo occurs in the Table 9 header.
  2. [Section 5.1.3 and Figure 3 caption] The spelling 'Possion' should be 'Poisson', and the Figure 3 caption duplicates the phrase 'the team in solution of the team in 1st place'.
  3. [Section 4.2] The model name 'YOLOv5mu' is unusual; the YOLOv5 family uses sizes n/s/m/l/x, so 'YOLOv5m' is likely intended.
  4. [CRediT authorship statement] The contributor 'Qiu Guan' is listed twice with 'Methodology'; remove the duplicate entry.
  5. [References [52] and [53]] References [52] and [53] are duplicate entries for the same Mixup paper; they should be consolidated into a single reference.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the paper reports direct empirical measurements on a new public dataset, with no fitted parameter, derivation, or prediction that reduces to its own inputs.

full rationale

The paper is a dataset-and-challenge summary rather than a derivation or modeling paper. Its central claims are (1) the VisAlgae 2023 dataset exists and is expert-annotated, (2) baseline detectors score a certain mAP50:95 on an 8:2 split of the training data, and (3) the top-10 leaderboard scores on the private test set are as listed in Table 3. None of these claims is defined in terms of the paper's own conclusions. The baseline numbers are direct measurements on a validation split; the leaderboard numbers are direct measurements by the Tianchi platform on the test set with private annotations. There is no parameter fitted to a subset and then renamed a prediction, no uniqueness theorem invoked from the authors' prior work, and no ansatz smuggled in via citation. The self-citations to [56, 57] are used only for details of the image-acquisition platform and prior challenge data, which are not load-bearing for the numerical results reported here; the dataset itself is released publicly and verified independently by expert annotators, so the citation is not the evidence for the dataset's properties. The protocol features that a skeptic might question — up to five submissions per day with automatic overwriting on a public leaderboard during a month-long window (Section 3.2), and the fact that the 'lightweight beats larger' baseline comparison in Section 4.2 is computed on the validation split rather than the private test set — are genuine correctness or generalizability risks, but they are not circularity. They concern whether the reported scores generalize or whether leaderboard probing inflated the top-10 results, not whether any claim is equivalent by construction to its inputs. Per the hard rules, concerns about benchmark validity without an exhibited definitional reduction belong to correctness risk, not circularity. Accordingly, no circular step can be quoted and exhibited, and the appropriate finding is no significant circularity with score 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claims rest on dataset annotation quality, split independence, leaderboard integrity, and comparability of baseline and test evaluations. None of these assumptions are independently verified in the manuscript, so the benchmark results should be treated as plausible but unproven until the dataset and split are independently audited.

assumptions (4)
  • domain assumption Domain experts correctly identified and annotated all six algae species in every image.
    Section 3.1.2 says images were 'authenticated by domain experts' and manually annotated with LabelImg, but no inter-annotator agreement or QA metrics are reported.
  • domain assumption The 700/300 train and test splits contain no near-duplicate images, despite both being frames from the same video sequences.
    Section 3.1.1 claims the sets are 'independent of each other without duplicate images', but temporally adjacent frames in video are often nearly identical; no frame-similarity analysis is provided.
  • domain assumption Public leaderboard scores are not inflated by repeated submissions probing the test set.
    Section 3.2 allows up to five submissions per day with automatic overwriting, which permits iterative test-set fitting; the paper does not analyze this risk.
  • domain assumption Baseline results on the 8:2 training/validation split are representative of the official test set.
    Section 4.1 splits the training set into train/val (8:2), while the challenge uses a separate 300-image test set; no test-set baseline scores are reported, so the baseline comparisons are not directly comparable to leaderboard scores.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VisAlgae 2023: A Dataset and Challenge for Algae Detection in Microscopy Images." pith.science (2026). https://pith.science/paper/YAVIL35T

@misc{pith2026250520687,
  author       = {Pith},
  title        = {Pith review of: VisAlgae 2023: A Dataset and Challenge for Algae Detection in Microscopy Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YAVIL35T}},
  note         = {Machine review of arXiv:2505.20687}
}
read the original abstract

Microalgae, vital for ecological balance and economic sectors, present challenges in detection due to their diverse sizes and conditions. This paper summarizes the second "Vision Meets Algae" (VisAlgae 2023) Challenge, aiming to enhance high-throughput microalgae cell detection. The challenge, which attracted 369 participating teams, includes a dataset of 1000 images across six classes, featuring microalgae of varying sizes and distinct features. Participants faced tasks such as detecting small targets, handling motion blur, and complex backgrounds. The top 10 methods, outlined here, offer insights into overcoming these challenges and maximizing detection accuracy. This intersection of algae research and computer vision offers promise for ecological understanding and technological advancement. The dataset can be accessed at: https://github.com/juntaoJianggavin/Visalgae2023/.

Figures

Figures reproduced from arXiv: 2505.20687 by the authors.

Figure 1
Figure 1. The statistics of annotations for each algal class are as follows: (a)Dataset images of algae in microfluidic channels; (b) object crops of each class; (c) the number of objects for each class; (d) the average aspect ratio and the ratio of their area to the entire image. "W/H" represents the aspect ratio, and "Ratio" represents the proportion of the area to the entire image. 3.2. Challenge Setup The testing phase wa… view at source ↗
Figure 2
Figure 2. The performance analysis of different baseline models in the algal classification task is as follows: (a)-(f) mAP50:95 on six algal classes: (Platymonas, Chlorella, Dunaliella salina, Effrenium, Porphyridium, Haematococcus)(g) Average performance across all classes. (h) Average per-class performance across all models. (i) Heatmap of mAP50:95 for each model and class. 10th place are shown in [PITH_FULL_IMAGE:figures… view at source ↗
Figure 3
Figure 3. Results of images after different augmentation methods from the solution of the team in solution of the team in 1 st place queue. The queue maintains a maximum capacity limit; once exceeded, older targets are randomly removed to preserve constant size. This mechanism accelerates target retrieval, thereby improving augmentation efficiency. Theoretically, the queue can cyclically store all algal targets across the dat… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The architecture of the 4th-place team improved YOLOv5 with CBAM and Transformer encoder blocks [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: The RepCSPLayer architecture in the solution of the team in 1st place Attention Mechanism Modules: To enhance feature representation, some teams incorporated attention-related modules. CBAM (Convolutional Block Attention Module) [48], an attention mechanism, weights fe…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 37 canonical work pages

  1. [1]

    Yolov4: Optimal speed and accuracy of object detection.arXiv:2004.10934

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

    arXiv 2020

  2. [2]

    Cascader-cnn:Delvingintohighqualityobjectdetection,in:ProceedingsoftheIEEEconferenceoncomputer vision and pattern recognition, pp

    Cai,Z.,Vasconcelos,N.,2018. Cascader-cnn:Delvingintohighqualityobjectdetection,in:ProceedingsoftheIEEEconferenceoncomputer vision and pattern recognition, pp. 6154–6162

    work page 2018

  3. [3]

    Cai,Z.,Vasconcelos,N.,2019.Cascader-cnn:Highqualityobjectdetectionandinstancesegmentation.IEEETransactionsonPatternAnalysis and Machine Intelligence , 1–1URL:http://dx.doi.org/10.1109/tpami.2019.2956516, doi:10.1109/tpami.2019.2956516

    arXiv 2019

  4. [4]

    Detection of microalgae objects based on the improved yolov3 model

    Cao, M., Wang, J., Chen, Y., Wang, Y., 2021. Detection of microalgae objects based on the improved yolov3 model. Environmental Science: Processes & Impacts 23, 1516–1530

    work page 2021

  5. [5]

    End-to-endobjectdetectionwithtransformers,in:ECCV

    Carion,N.,Massa,F.,Synnaeve,G.,Usunier,N.,Kirillov,A.,Zagoruyko,S.,2020. End-to-endobjectdetectionwithtransformers,in:ECCV

    work page 2020

  6. [6]

    MMDetection:Openmmlabdetectiontoolbox and benchmark

    Chen,K.,Wang,J.,Pang,J.,Cao,Y.,Xiong,Y.,Li,X.,Sun,S.,Feng,W.,Liu,Z.,Xu,J.,Zhang,Z.,Cheng,D.,Zhu,C.,Cheng,T.,Zhao,Q., Li,B.,Lu,X.,Zhu,R.,Wu,Y.,Dai,J.,Wang,J.,Shi,J.,Ouyang,W.,Loy,C.C.,Lin,D.,2019. MMDetection:Openmmlabdetectiontoolbox and benchmark. arXiv preprint arXiv:1906.07155

    arXiv 2019

  7. [7]

    Diffusiondet: Diffusion model for object detection

    Chen, S., Sun, P., Song, Y., Luo, P., 2022. Diffusiondet: Diffusion model for object detection. arXiv preprint arXiv:2211.09788

    arXiv 2022

  8. [8]

    Histograms of oriented gradients for human detection, in: 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR’05), Ieee

    Dalal, N., Triggs, B., 2005. Histograms of oriented gradients for human detection, in: 2005 IEEE computer society conference on computer vision and pattern recognition (CVPR’05), Ieee. pp. 886–893

    work page 2005

Show all 58 references
  1. [9]

    Bioindicators in marine waters: benthic diatoms as a tool to assess water quality from eutrophic to oligotrophic coastal ecosystems

    Desrosiers, C., Leflaive, J., Eulin, A., Ten-Hage, L., 2013. Bioindicators in marine waters: benthic diatoms as a tool to assess water quality from eutrophic to oligotrophic coastal ecosystems. Ecological indicators 32, 25–34

    2013

  2. [10]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al., 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929

    2020 arXiv

  3. [11]

    Data augmentation for object detection via controllable diffusion models, in: Proceedings of the IEEE/CVF winter conference on applications of computer vision, pp

    Fang, H., Han, B., Zhang, S., Zhou, S., Hu, C., Ye, W.M., 2024. Data augmentation for object detection via controllable diffusion models, in: Proceedings of the IEEE/CVF winter conference on applications of computer vision, pp. 1257–1266

    2024

  4. [12]

    Simple copy-paste is a strong data augmentation method for instance segmentation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Ghiasi, G., Cui, Y., Srinivas, A., Qian, R., Lin, T.Y., Cubuk, E.D., Le, Q.V., Zoph, B., 2021. Simple copy-paste is a strong data augmentation method for instance segmentation, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 2918–2928

    2021

  5. [13]

    Nas-fpn: Learning scalable feature pyramid architecture for object detection, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Ghiasi, G., Lin, T.Y., Le, Q.V., 2019. Nas-fpn: Learning scalable feature pyramid architecture for object detection, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 7036–7045

    2019

  6. [14]

    Fast r-cnn, in: Proceedings of the IEEE international conference on computer vision, pp

    Girshick, R., 2015. Fast r-cnn, in: Proceedings of the IEEE international conference on computer vision, pp. 1440–1448

    2015

  7. [15]

    Rich feature hierarchies for accurate object detection and semantic segmentation, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Girshick, R., Donahue, J., Darrell, T., Malik, J., 2014. Rich feature hierarchies for accurate object detection and semantic segmentation, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 580–587. 14

    2014

  8. [16]

    Deep residual learning for image recognition, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    He, K., Zhang, X., Ren, S., Sun, J., 2016. Deep residual learning for image recognition, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778

    2016

  9. [17]

    Coordinate attention for efficient mobile network design, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Hou, Q., Zhou, D., Feng, J., 2021. Coordinate attention for efficient mobile network design, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 13713–13722

    2021

  10. [18]

    Ultralytics yolov5

    Jocher, G., 2020. Ultralytics yolov5. URL:https://github.com/ultralytics/yolov5, doi:10.5281/zenodo.3908559

    2020 doi

  11. [19]

    Ultralytics yolov8

    Jocher, G., Chaurasia, A., Qiu, J., 2023. Ultralytics yolov8. URL:https://github.com/ultralytics/ultralytics

    2023

  12. [20]

    Ultralytics yolo11

    Jocher, G., Qiu, J., 2024. Ultralytics yolo11. URL:https://github.com/ultralytics/ultralytics

    2024

  13. [21]

    Deep learning

    LeCun, Y., Bengio, Y., Hinton, G., 2015. Deep learning. nature 521, 436–444

    2015

  14. [22]

    Yolov6 v3.0: A full-scale reloading

    Li, C., Li, L., Geng, Y., Jiang, H., Cheng, M., Zhang, B., Ke, Z., Xu, X., Chu, X., 2023. Yolov6 v3.0: A full-scale reloading. arXiv:2301.05586

    2023 arXiv

  15. [23]

    An improved algae-yolo model based on deep learning for object detection of ocean microalgae considering aquacultural lightweight deployment

    Liu, D., Wang, P., Cheng, Y., Bi, H., 2022. An improved algae-yolo model based on deep learning for object detection of ocean microalgae considering aquacultural lightweight deployment. Frontiers in Marine Science 9, 2378

    2022

  16. [24]

    Path aggregation network for instance segmentation, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Liu, S., Qi, L., Qin, H., Shi, J., Jia, J., 2018. Path aggregation network for instance segmentation, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 8759–8768

    2018

  17. [25]

    Ssd: Single shot multibox detector

    Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.Y., Berg, A.C., 2016. Ssd: Single shot multibox detector. ECCV

    2016

  18. [26]

    Swin transformer: Hierarchical vision transformer using shifted windows, in: Proceedings of the IEEE/CVF international conference on computer vision, pp

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., Guo, B., 2021. Swin transformer: Hierarchical vision transformer using shifted windows, in: Proceedings of the IEEE/CVF international conference on computer vision, pp. 10012–10022

    2021

  19. [27]

    Detrs beat yolos on real-time object detection

    Lv, W., Xu, S., Zhao, Y., Wang, G., Wei, J., Cui, C., Du, Y., Dang, Q., Liu, Y., 2023. Detrs beat yolos on real-time object detection. arXiv preprint arXiv:2304.08069

    2023 arXiv

  20. [28]

    Rtmdet: An empirical study of designing real-time object detectors

    Lyu, C., Zhang, W., Huang, H., Zhou, Y., Wang, Y., Liu, Y., Zhang, S., Chen, K., 2022. Rtmdet: An empirical study of designing real-time object detectors. arXiv preprint arXiv:2212.07784

    2022 arXiv

  21. [29]

    Science of the Total Environment 802, 149800

    O’Neill,E.A.,Rowan,N.J.,2022.Microalgaeasanaturalecologicalbioindicatorforthesimplereal-timemonitoringofaquaculturewastewater qualityincludingprovisionforassessingimpactofextremesinclimatevariance–acomparativecasestudyfromtherepublicofireland. Science of the Total Environment ...

    2022

  22. [30]

    Onthe integration of self-attentionand convolution, in:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Pan, X.,Ge, C., Lu, R.,Song, S., Chen, G.,Huang, Z., Huang, G.,2022. Onthe integration of self-attentionand convolution, in:Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 815–825

    2022

  23. [31]

    Microalgae detection using a deep learning object detection algorithm, yolov3

    Park, J., Baek, J., You, K., Nam, S.W., Kim, J., 2021. Microalgae detection using a deep learning object detection algorithm, yolov3. Journal of Korean Society on Water Environment 37, 275–285

    2021

  24. [32]

    Youonlylookonce:Unified,real-timeobjectdetection,in:ProceedingsoftheIEEE conference on computer vision and pattern recognition, pp

    Redmon,J.,Divvala,S.,Girshick,R.,Farhadi,A.,2016. Youonlylookonce:Unified,real-timeobjectdetection,in:ProceedingsoftheIEEE conference on computer vision and pattern recognition, pp. 779–788

    2016

  25. [33]

    Yolo9000: better, faster, stronger, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp

    Redmon, J., Farhadi, A., 2017. Yolo9000: better, faster, stronger, in: Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 7263–7271

    2017

  26. [34]

    Yolov3: An incremental improvement

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

    2018 arXiv

  27. [35]

    Faster r-cnn: Towards real-time object detection with region proposal networks

    Ren, S., He, K., Girshick, R., Sun, J., 2015. Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems 28

    2015

  28. [36]

    Weighted boxes fusion: Ensembling boxes from different object detection models

    Solovyev, R., Wang, W., Gabruseva, T., 2021. Weighted boxes fusion: Ensembling boxes from different object detection models. Image and Vision Computing , 1–6

    2021

  29. [37]

    Commercial applications of microalgae

    Spolaore, P., Joannis-Cassan, C., Duran, E., Isambert, A., 2006. Commercial applications of microalgae. Journal of bioscience and bioengineering 101, 87–96

    2006

  30. [38]

    Efficientdet: Scalable and efficient object detection, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Tan, M., Pang, R., Le, Q.V., 2020. Efficientdet: Scalable and efficient object detection, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 10781–10790

    2020

  31. [39]

    Yolov12: Attention-centric real-time object detectors

    Tian, Y., Ye, Q., Doermann, D., 2025. Yolov12: Attention-centric real-time object detectors. arXiv preprint arXiv:2502.12524

    2025 arXiv

  32. [40]

    Effective data augmentation with diffusion models

    Trabucco, B., Doherty, K., Gurinas, M., Salakhutdinov, R., 2023. Effective data augmentation with diffusion models. arXiv preprint arXiv:2302.07944

    2023 arXiv

  33. [41]

    Rapid object detection using a boosted cascade of simple features, in: Proceedings of the 2001 IEEE computer society conference on computer vision and pattern recognition

    Viola, P., Jones, M., 2001. Rapid object detection using a boosted cascade of simple features, in: Proceedings of the 2001 IEEE computer society conference on computer vision and pattern recognition. CVPR 2001, Ieee. pp. I–I

    2001

  34. [42]

    Yolov10: Real-time end-to-end object detection

    Wang, A., Chen, H., Liu, L., et al., 2024. Yolov10: Real-time end-to-end object detection. arXiv preprint arXiv:2405.14458

    2024 arXiv

  35. [43]

    YOLOv7:Trainablebag-of-freebiessetsnewstate-of-the-artforreal-timeobjectdetectors

    Wang,C.Y.,Bochkovskiy,A.,Liao,H.Y.M.,2022. YOLOv7:Trainablebag-of-freebiessetsnewstate-of-the-artforreal-timeobjectdetectors. arXiv preprint arXiv:2207.02696

    2022 arXiv

  36. [44]

    YOLOv9: Learning what you want to learn using programmable gradient information

    Wang, C.Y., Liao, H.Y.M., 2024. YOLOv9: Learning what you want to learn using programmable gradient information

    2024

  37. [45]

    Carafe:Content-awarereassemblyoffeatures,in:ProceedingsoftheIEEE/CVF International Conference on Computer Vision (ICCV)

    Wang,J.,Chen,K.,Xu,R.,Liu,Z.,Loy,C.C.,Lin,D.,2019. Carafe:Content-awarereassemblyoffeatures,in:ProceedingsoftheIEEE/CVF International Conference on Computer Vision (ICCV)

    2019

  38. [46]

    Internimage: Exploring large-scale vision foundation models with deformable convolutions, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Wang, W., Dai, J., Chen, Z., Huang, Z., Li, Z., Zhu, X., Hu, X., Lu, T., Lu, L., Li, H., et al., 2023. Internimage: Exploring large-scale vision foundation models with deformable convolutions, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognitio...

    2023

  39. [47]

    Convnext v2: Co-designing and scaling convnets with masked autoencoders, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp

    Woo, S., Debnath, S., Hu, R., Chen, X., Liu, Z., Kweon, I.S., Xie, S., 2023. Convnext v2: Co-designing and scaling convnets with masked autoencoders, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 16133–16142

    2023

  40. [48]

    Cbam: Convolutional block attention module, in: Proceedings of the European conference on computer vision (ECCV), pp

    Woo, S., Park, J., Lee, J.Y., Kweon, I.S., 2018. Cbam: Convolutional block attention module, in: Proceedings of the European conference on computer vision (ECCV), pp. 3–19

    2018

  41. [49]

    Aggregated residual transformations for deep neural networks.arXiv:1611.05431

    Xie, S., Girshick, R., Dollár, P., Tu, Z., He, K., 2017. Aggregated residual transformations for deep neural networks.arXiv:1611.05431

    2017 arXiv

  42. [50]

    Yolox model-based object detection for microalgal bioprocess

    Yan, H., Peng, X., Chen, C., Xia, A., Huang, Y., Zhu, X., Zhu, X., Liao, Q., 2023. Yolox model-based object detection for microalgal bioprocess. Algal Research , 103178

    2023

  43. [51]

    Focal modulation networks

    Yang, J., Li, C., Dai, X., Gao, J., 2022. Focal modulation networks. Advances in Neural Information Processing Systems 35, 4203–4217. 15

    2022

  44. [53]

    mixup: Beyond empirical risk minimization.arXiv:1710.09412

    Zhang, H., Cisse, M., Dauphin, Y.N., Lopez-Paz, D., 2018. mixup: Beyond empirical risk minimization.arXiv:1710.09412

    2018 arXiv

  45. [54]

    Enhancingrareobjectdetectiononroadwaysthroughconditionaldiffusion models for data augmentation

    Zhang,H.,Hu,Y.,Qian,Z.,Sha,J.,Xie,M.,Wan,Y.,Liu,P.,2024. Enhancingrareobjectdetectiononroadwaysthroughconditionaldiffusion models for data augmentation. IEEE Transactions on Intelligent Transportation Systems

    2024

  46. [55]

    Computer vision meets microfluidics: a label-free method for high-throughput cell analysis

    Zhou, S., Chen, B., Fu, E.S., Yan, H., 2023a. Computer vision meets microfluidics: a label-free method for high-throughput cell analysis. Microsystems & Nanoengineering 9, 116

  47. [56]

    A microfluidic microalgae detection system for cellular physiological response based on an object detection algorithm

    Zhou, S., Chen, T., Fu, E.S., Zhou, T., Shi, L., Yan, H., 2024. A microfluidic microalgae detection system for cellular physiological response based on an object detection algorithm. Lab on a Chip 24, 2762–2773

    2024

  48. [57]

    Vision meets algae: A novel way for microalgae recognization and health monitor

    Zhou, S., Jiang, J., Hong, X., Fu, P., Yan, H., 2023b. Vision meets algae: A novel way for microalgae recognization and health monitor. Frontiers in Marine Science 10, 1105545

  49. [58]

    Deformable detr: Deformable transformers for end-to-end object detection, in: International Conference on Learning Representations

    Zhu, X., Su, W., Lu, L., Li, B., Wang, X., Dai, J., 2021. Deformable detr: Deformable transformers for end-to-end object detection, in: International Conference on Learning Representations. URL:https://openreview.net/forum?id=gZ9hCDWe6ke

    2021

  50. [59]

    Detrs with collaborative hybrid assignments training, in: Proceedings of the IEEE/CVF international conference on computer vision, pp

    Zong, Z., Song, G., Liu, Y., 2023. Detrs with collaborative hybrid assignments training, in: Proceedings of the IEEE/CVF international conference on computer vision, pp. 6748–6758. 16

    2023

Pith tools

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