Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

FishDet-M: A Unified Large-Scale Benchmark for Robust Fish Detection and CLIP-Guided Model Selection in Diverse Aquatic Visual Domains

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read FishDet-M merges 13 public fish datasets into a 105,556-image COCO-style benchmark, benchmarks 28 detectors, and reports YOLO12x at 0.491 mAP with a CLIP-powered per-image model selector.

desk verdict A genuinely useful fish-detection benchmark whose leaderboard currently rests on under-specified training and unverified split disjointness; worth reviewing carefully, but not accepting as-is. read the letter →

arxiv 2507.17859 v1 pith:YQPT6TET submitted 2025-07-23 cs.CV cs.RO

classification cs.CVcs.RO
keywords fishdetectionunderwatervisionobjectbenchmarkdatasetharmonizationYOLOCLIPmodelselectiondomaingeneralizationCOCOannotations
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

FishDet-M is the paper's attempt to give underwater fish detection a single, standardized yardstick: it merges 13 public datasets covering marine, brackish, occluded, and aquarium scenes into 105,556 images and 296,885 COCO-style fish instances, then trains and evaluates 28 detectors on the same splits. The headline result is a reproducible leaderboard in which YOLO12x reaches 0.491 mAP, with the YOLO family consistently ahead of R-CNN and DETR-style models, and where lighter YOLO variants offer useful speed-accuracy trade-offs. The paper also proposes a CLIP-based selector that reads each image, scores it against text prompts, and routes it to the model with the highest semantic alignment, reaching 0.444 mAP at 80 FPS without ensembling. If the benchmark holds up, the field gains a common corpus and protocol that make cross-study comparisons of underwater detectors meaningful for the first time.

What carries the argument

The central object is the benchmark itself: a harmonization pipeline that unifies 13 datasets into a single COCO-style corpus with 296,885 fish instances, deduplicated frames, and a source-aware stratified split, so that every model is trained and tested on identical data. The second mechanism is the CLIP-based model router, which embeds an input image and compares it against text prompts describing each of the 28 candidate detectors, then selects the highest-scoring model for that image: a zero-shot, per-image routing rule that needs no ensemble computation. The third piece is the evaluation protocol itself—mAP at multiple IoU thresholds, scale-specific AP for small, medium, and large objects, plus latency and parameter counts—which is what turns the merged corpus into a reproducible leaderboard.

What would settle it

Run a near-duplicate detection pass, such as perceptual hashing, over the full FishDet-M corpus; if any test-set image has a near-duplicate in the training set, the reported mAP numbers are optimistic and the benchmark's core claim fails. Alternatively, retrain YOLO12x and Deformable-DETR on the released splits with the stated defaults; if the ranking flips materially, by more than a few mAP points, or YOLO12x no longer leads, the leaderboard is not reproducible as released.

Watch

Extended reading notes

Core claim

Built from thirteen public sources, FishDet-M harmonizes bounding boxes and segmentation masks into a species-agnostic 'fish' category in COCO format, with a source-aware stratified split of 80% training, 10% validation, and 10% test. On this benchmark, the paper reports that the YOLO family dominates: YOLO12x scores 0.491 mAP, YOLO12l 0.487, and YOLO11l 0.484, while transformer-based detectors such as DETR, Deformable-DETR, and RT-DETR-l trail at 0.317-0.390 mAP and show weaker small-object performance. The CLIP-guided selector (FishDet-M-CLIP) picks per-image models by measuring similarity between the image embedding and model-specific prompt texts; it achieves 0.444 mAP at 80.28 FPS, outperforming most traditional detectors but not the top fixed YOLO models. Models trained on the merged data also transfer to a 1,500-image external dataset, with YOLO variants reaching roughly 0.63 mAP, which the paper reads as evidence of domain-generalizable feature learning.

Load-bearing premise

Every model comparison in the paper depends on the test split being genuinely clean, with no duplicate or near-identical frames leaking from the training data of the thirteen merged sources, and on the 1,500-image external test set being an honest independent sample; the paper asserts deduplication but offers no quantitative leakage check, and the external set is an unvalidated single-source upload cited only as 'test'.

Editorial extensions

If this is right

  • A common 105,556-image corpus with fixed splits lets any future fish detector be compared directly against the 28-model leaderboard, ending the practice of each paper reporting on its own private subset.
  • The reported YOLO dominance suggests that, for turbid underwater scenes, attention-centric convolutional detectors are currently a better default than DETR-style transformers, especially where small fish matter.
  • The CLIP routing result implies that a vision-language model can serve as a cheap arbitration layer that picks detectors per image, hinting at deployment pipelines that adapt model choice to water conditions.
  • If the generalization results reproduce, training on merged diverse data transfers better to unseen sources than single-domain training, favoring aggregated benchmarks for field deployment.
  • Scale-aware metrics (AP_S, AP_M, AP_L) reveal that small-object performance is the principal differentiator among models, pointing to where future underwater detectors should invest capacity.

Reading between the lines

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

  • The paper does not analyze which source dataset drives YOLO's advantage; a next test would be to ablate the 13 sources and check whether top models depend on the largest contributor's 94,532 images or on the harder turbid subsets, which would tell practitioners whether smaller diverse sets could replace the full collection.
  • Because the external generalization set is a single 1,500-image upload cited only as 'test', with no documented provenance or license, the claim of domain-generalizable learning should be re-tested on two or three independently curated underwater test sets before being relied on.
  • The CLIP selector's 0.444 mAP is below the best fixed YOLO; an untested extension is to pair the router with confidence-weighted voting among the top-k selected models, which could recover the gap while keeping latency under ensemble cost.
  • The species-agnostic 'fish' label means the benchmark cannot yet test class-specific confusion, such as camouflage against background species; enriching a subset with species tags would let the benchmark address fine-grained questions.
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. This paper introduces FishDet-M, a unified fish detection benchmark constructed by harmonizing 13 public underwater datasets into COCO-style annotations, totaling 105,556 images and 296,885 instances split into training, validation, and test sets. The authors benchmark 28 object detectors, mostly YOLO variants, R-CNN-style models, and DETR-based models, reporting mAP, mAP@50, mAP@75, scale-wise AP, latency, FPS, and parameter counts. They also propose a CLIP-based model selection mechanism that routes each input image to one of the trained detectors based on text-prompt similarity, and they evaluate generalization on a separate 1,500-image test set. The main claims are that FishDet-M is the largest unified fish detection benchmark and that it provides a standardized, reproducible platform for cross-domain evaluation.

Significance. If the dataset construction and evaluation protocols are sound, FishDet-M would be a useful community resource: it consolidates fragmented fish-detection data, provides a common annotation format, evaluates a broad set of modern detectors under one protocol, and promises public release of data, code, and checkpoints. The scale of the harmonization effort and the breadth of the detector comparison are real contributions, and the CLIP-guided selection idea is a plausible direction for adaptive deployment. However, the significance of the benchmark hinges on the reliability of the reported leaderboard and on the absence of train/test leakage, and these points are currently not demonstrated. The paper also claims segmentation masks that do not appear to be described or released, which weakens the data-record contribution as written.

major comments (5)
  1. [Section IV, Technical Validation; Table VI] The only training description is that models were trained using "defualt hyperparameters." The manuscript does not report training epochs, batch size, input resolution, optimizer and learning-rate schedule, weight decay, augmentation settings, random seeds, number of runs, or the checkpoint-selection rule (last epoch versus best validation mAP). Because Table VI is the central deliverable of the benchmark, this omission makes the leaderboard non-reproducible and leaves open the possibility that the small margins at the top (YOLO12x 0.491 vs. YOLO12l 0.487 vs. YOLO11l 0.484) reflect per-framework default configurations rather than model quality. Please specify the complete protocol for each codebase, including the exact defaults used in Ultralytics, MMDetection, and SuperGradients, and report validation-based model selection details.
  2. [Section III, Partitioning Strategy and Harmonization] The paper states that "non-representative or duplicate frames were removed" to prevent data leaks, but it provides no quantitative analysis showing that the training and test splits are disjoint at the image or near-duplicate level. Several source datasets are video-derived (Brackish-MOT, Fish Video, FISH-video, Fish4Knowledge, FishNet), and random frame-level splits can place near-identical frames in both training and test partitions. Given how tightly packed the top of the leaderboard is, even a small number of leaked near-duplicates can change rankings and undermine the cross-domain generalization claims. Please add a per-source analysis of exact and near-duplicate overlap between the train and test partitions, for example using perceptual hashing or feature similarity, and report the number of removed or remaining near-duplicates.
  3. [Section V.B and Table VII] The external generalization test set is described only as 1,500 unseen images from a different source [95], and reference [95] is a Roboflow dataset page from the user "test." No license, provenance, composition, annotation procedure, or overlap analysis with the 13 FishDet-M source datasets is provided. If the external set overlaps with FishDet-M or with any constituent dataset, the generalization results in Table VII become circular. Please document this dataset fully, including its origin, species and environmental distribution, annotation quality, license, and a quantitative overlap check with FishDet-M and its sources, or remove it from the generalization claim.
  4. [Section V.C and Table VI] The CLIP-based model selection system is under-specified. The prompt used for selection is said to be in the supplementary material, but it should be in the main text or in an attached supplement. More importantly, the paper does not define how the reported FishDet-M-CLIP mAP of 0.444 is computed, including whether the selected detector uses its individual confidence threshold and whether evaluation is done with the same 11,809 test images and identical evaluation code. There is also no baseline comparison such as random model selection, always-selecting-the-best-fixed-model, or an oracle selector, so the reader cannot determine whether the CLIP similarity scores provide real routing signal or merely reflect a preference for larger YOLO variants. Please add these details and baselines.
  5. [Abstract, Section I, Section IV, Table IV] The abstract and introduction claim that FishDet-M provides "COCO-style annotations with both bounding boxes and segmentation masks," but Section IV's harmonization description only discusses bounding boxes, and Table IV lists bounding boxes but not masks. Either the masks are not actually included, in which case the claim should be corrected, or they are included, in which case the mask harmonization process, mask statistics, and release status need to be described. This discrepancy is load-bearing for the data-record contribution and should be resolved.
minor comments (5)
  1. [Section IV, Technical Validation] The word "defualt" should be "default".
  2. [Section V.A.3 and Section V.D] The reported FishDataset mAP is inconsistent: Section V.A.3 says YOLO12x scored 0.371 on FishDataset, while Section V.D says the range goes down to 0.359 on FishDataset. Please reconcile these numbers.
  3. [Section III and Section V.E] The reported mean image resolution is 715×468 in Section III but 711×465 in Section V.E; these values should be made consistent.
  4. [Table VI] The FishDet-M-CLIP row has no parameter count and does not clarify whether the reported inference time includes both the CLIP encoder and the selected detector. Please specify the exact components and overhead breakdown.
  5. [Section VI] There is a typo in "such asvYOLOv12x" in the limitations section, which should read "such as YOLOv12x."

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; benchmark construction and model evaluation are self-contained.

full rationale

FishDet-M is a benchmark paper, and its derivation chain is straightforward: merge 13 public datasets into a unified COCO-style format, split them into train/validation/test, train 28 off-the-shelf detectors with default hyperparameters, and report metrics on the held-out test split. No parameter, threshold, or fitted constant is derived from the test labels and then renamed as a prediction; the mAP values in Table VI are standard pycocotools evaluations of independently trained models. The CLIP-guided model selector (FishDet-M-CLIP) is evaluated on the same fixed test set like any other candidate, and the paper does not claim to fit its prompts to test performance; it reports a lower mAP than the top fixed YOLO models, which is an honest outcome rather than a forced one. The generalization test in Table VII uses a separate 1,500-image corpus (ref [95]), and although that corpus is thinly documented, the paper does not reduce the generalization claim to the FishDet-M test set by construction. The self-citations (refs [23], [31], [54], [67], [96]) are related-work and supporting-context citations, not load-bearing justifications for the benchmark's validity or for any derived result. The stated concerns about train/test disjointness, near-duplicate frames, and the provenance of the external Roboflow 'test' dataset are leakage and correctness risks, not circularity: even if leakage occurred, it would inflate scores experimentally rather than make the output equivalent to the input by definition. No equation equates a reported prediction with a fitted input, and no uniqueness theorem or ansatz is imported from the authors' prior work to force a choice. Therefore the paper receives a score of 0, indicating no significant circularity.

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

The paper's central contribution is a dataset and an evaluation, not a new theory. Its hidden assumptions are about data hygiene and label consistency rather than physical laws. The lack of leakage analysis, the reliance on default training settings across different frameworks, and the unvalidated external test set are the main unspecified premises.

assumptions (3)
  • domain assumption All 13 source datasets can be meaningfully merged into a single species-agnostic 'fish' category without losing the intended detection semantics.
    Section IV states that all bounding boxes were harmonized into a unified, species-agnostic fish category. If datasets differ in what counts as a fish (e.g., debris datasets like TrashCan 1.0, or multi-class datasets like AAS), the merged annotations may encode inconsistent label semantics.
  • domain assumption The 10% test split is fully disjoint from training and validation after deduplication.
    Section III describes a source-aware stratified split and Section IV says duplicate frames were removed to prevent data leaks, but no quantitative leakage analysis or duplicate-removal verification is provided. Near-duplicate frames across sources could inflate scores.
  • domain assumption Default hyperparameters give comparable training quality across frameworks (Ultralytics, MMDetection, Supergradients).
    Section IV states that models were trained using 'defualt hyperparameters'. Since each framework has different defaults for epochs, batch size, LR schedule, and augmentation, observed accuracy gaps may partly reflect framework defaults rather than detector architecture.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FishDet-M: A Unified Large-Scale Benchmark for Robust Fish Detection and CLIP-Guided Model Selection in Diverse Aquatic Visual Domains." pith.science (2026). https://pith.science/paper/YQPT6TET

@misc{pith2026250717859,
  author       = {Pith},
  title        = {Pith review of: FishDet-M: A Unified Large-Scale Benchmark for Robust Fish Detection and CLIP-Guided Model Selection in Diverse Aquatic Visual Domains},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YQPT6TET}},
  note         = {Machine review of arXiv:2507.17859}
}
abstract

Accurate fish detection in underwater imagery is essential for ecological monitoring, aquaculture automation, and robotic perception. However, practical deployment remains limited by fragmented datasets, heterogeneous imaging conditions, and inconsistent evaluation protocols. To address these gaps, we present \textit{FishDet-M}, the largest unified benchmark for fish detection, comprising 13 publicly available datasets spanning diverse aquatic environments including marine, brackish, occluded, and aquarium scenes. All data are harmonized using COCO-style annotations with both bounding boxes and segmentation masks, enabling consistent and scalable cross-domain evaluation. We systematically benchmark 28 contemporary object detection models, covering the YOLOv8 to YOLOv12 series, R-CNN based detectors, and DETR based models. Evaluations are conducted using standard metrics including mAP, mAP@50, and mAP@75, along with scale-specific analyses (AP$_S$, AP$_M$, AP$_L$) and inference profiling in terms of latency and parameter count. The results highlight the varying detection performance across models trained on FishDet-M, as well as the trade-off between accuracy and efficiency across models of different architectures. To support adaptive deployment, we introduce a CLIP-based model selection framework that leverages vision-language alignment to dynamically identify the most semantically appropriate detector for each input image. This zero-shot selection strategy achieves high performance without requiring ensemble computation, offering a scalable solution for real-time applications. FishDet-M establishes a standardized and reproducible platform for evaluating object detection in complex aquatic scenes. All datasets, pretrained models, and evaluation tools are publicly available to facilitate future research in underwater computer vision and intelligent marine systems.

Figures

Figures reproduced from arXiv: 2507.17859 by the authors.

Figure 1
Figure 1. Interactive application interface for comparing detection outputs on FishDet-M. Supports multiple models, bounding box [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Distribution of object-level statistics across FishDet [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Visual property distributions. (a) Image resolution. (b) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 7
Figure 7. Figure 7: CLIP-based model selection pipeline. The input image [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: F1 score and Precision-Recall curves for 28 object [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Object detection results from 28 models across four challenging underwater images. Each row shows the original image [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Frequency of model selections based on CLIP similarity [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]
Figure 11
Figure 11. Figure 11: Distribution of CLIP similarity scores across all [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Energy Constrained Hierarchical Underwater Monitoring via Local Multi-Agent RAG

    cs.IR 2026-07 conditional novelty 6.0 of 10

    A master–satellite edge station pairs MAX78000/02 always-on visual/acoustic sentinels with selective Jetson multimodal RAG, local species ID, and multi-agent reporting to cut energy and uplink cost.

Reference graph

Works this paper leans on

100 extracted references · 74 canonical work pages · cited by 1 Pith paper

  1. [95]

    Fcos: Fully convolutional one- stage object detection,

    Z. Tian, C. Shen, H. Chen, and T. He, “Fcos: Fully convolutional one- stage object detection,” pp. 9626–9635, Oct 2019

  2. [1]

    Fish- finder: A robust small target detection method for aquaculture fish in low-quality underwater images,

    L. Liu, J. Wu, H. Zhao, H. Kong, T. Zheng, B. Qu, and H. Yu, “Fish- finder: A robust small target detection method for aquaculture fish in low-quality underwater images,” Journal of Fish Biology , vol. 106, no. 3, pp. 908–920, 2025

  3. [2]

    Real-time fish detection in complex backgrounds using probabilistic background modelling,

    A. Salman, S. Maqbool, A. H. Khan, A. Jalal, and F. Shafait, “Real-time fish detection in complex backgrounds using probabilistic background modelling,” Ecological Informatics, vol. 51, pp. 44–51, 2019

  4. [3]

    Underwater object detection: architectures and algorithms – a comprehensive review,

    S. Fayaz, S. A. Parah, and G. J. Qureshi, “Underwater object detection: architectures and algorithms – a comprehensive review,” Multimedia Tools and Applications, vol. 81, no. 15, pp. 20 871–20 916, June 2022

  5. [4]

    Fish detection under occlusion using modified you only look once v8 integrating real- time detection transformer features,

    E. Li, Q. Wang, J. Zhang, W. Zhang, H. Mo, and Y . Wu, “Fish detection under occlusion using modified you only look once v8 integrating real- time detection transformer features,” Applied Sciences, vol. 13, no. 23, 2023

  6. [5]

    Automated underwater fish species recognition using deep learning-based techniques,

    P. Anantha Prabha, S. Sachin, U. Srinithish, M. Deva Priya, and S. Karthick, “Automated underwater fish species recognition using deep learning-based techniques,” in Proceedings of International Conference on Recent Trends in Computing , R. P. Mahapatra, S. K. Peddoju, S. Roy, and P. Parwekar, Eds. Singapore: Springer Nature Singapore, 2024, pp. 807–815

  7. [6]

    Underwater fish detection and classification using deep learning,

    V . Pagire and A. Phadke, “Underwater fish detection and classification using deep learning,” in 2022 International Conference on Intelligent Controller and Computing for Smart Power (ICICCSP) , 2022, pp. 1–4

  8. [7]

    Accelerating species recognition and labelling of fish from underwater video with machine-assisted deep learning,

    D. Marrable, K. Barker, S. Tippaya, M. Wyatt, S. Bainbridge, M. Stowar, and J. Larke, “Accelerating species recognition and labelling of fish from underwater video with machine-assisted deep learning,” Frontiers in Marine Science , vol. V olume 9 - 2022, 2022

Show all 100 references
  1. [8]

    Depondfi’23 challenge on real-time pond environ- ment: Methods and results,

    A. Sasithradevi, R. Suganya, P. Prakash, S. Mohamed Mansoor Roomi, M. Vijayalakshmi, S. Nathan, P. Kasthuri, J. Persiya, L. Brighty Ebenezer, S. Jain, S. Verma, S. Balasubramanian, M. Sai Subramaniam, T. Sai Sri- ram, M. Pranav Phanindra Sai, C. Raj, A. Yadav, R. Payak, S. Pau...

  2. [9]

    A benchmark dataset and ensemble yolo method for enhanced underwater fish detection,

    V . Mohankumar and S. Anbalagan, “A benchmark dataset and ensemble yolo method for enhanced underwater fish detection,” ETRI Journal . [Online]. Available: https://doi.org/10.4218/etrij.2024-0383

  3. [10]

    A feature- enhanced and adaptive routing framework for fish school detection on auvs for degraded underwater imaging environments,

    Y . Jiang, Y . Wang, Y . Zhang, Q. Guo, M. Zhao, and H. Qin, “A feature- enhanced and adaptive routing framework for fish school detection on auvs for degraded underwater imaging environments,” IEEE Internet of Things Journal, vol. 11, no. 10, pp. 18 335–18 350, May 2024

  4. [11]

    Fishdet-yolo: Enhanced underwater fish detection with richer gradient flow and long-range dependency capture through mamba-c2f,

    C. Yang, J. Xiang, X. Li, and Y . Xie, “Fishdet-yolo: Enhanced underwater fish detection with richer gradient flow and long-range dependency capture through mamba-c2f,” Electronics, vol. 13, no. 18, 2024

  5. [12]

    Yolov8-tf: Transformer-enhanced yolov8 for underwater fish species recognition with class imbalance handling,

    C. Shah, M. M. Nabi, S. Y . Alaba, I. A. Ebu, J. Prior, M. D. Campbell, R. Caillouet, M. D. Grossi, T. Rowell, F. Wallace, J. E. Ball, and R. Moorhead, “Yolov8-tf: Transformer-enhanced yolov8 for underwater fish species recognition with class imbalance handling,” Sensors, vol....

  6. [13]

    Intelligent deep learning based automated fish detection model for uwsn,

    M. A. Duhayyim, H. M. Alshahrani, F. N. Al-Wesabi, M. Alamgeer, A. M. Hilal, and M. A. Hamza, “Intelligent deep learning based automated fish detection model for uwsn,” Computers, Materials & Continua , vol. 70, no. 3, pp. 5871–5887, 2022

  7. [14]

    Deep fish: An approach to fish species identification through deep learning techniques,

    P. J. Reddy, M. Malathi, and A. N. Julaiha, “Deep fish: An approach to fish species identification through deep learning techniques,” in Emerging Trends in Expert Applications and Security , V . S. Rathore, V . Piuri, R. Babo, and V . Tiwari, Eds. Singapore: Springer Nature Si...

  8. [15]

    A 3d occlusion tracking model of the underwater fish targets,

    M. Jia-Fa, X. Gang, S. Wei-Guo, and X. Liu, “A 3d occlusion tracking model of the underwater fish targets,” in 2015 IEEE International Conference on Electro/Information Technology (EIT) , May 2015, pp. 082–086

  9. [16]

    A multitask model for realtime fish detection and segmentation based on YOLOv5,

    Q. Liu, X. Gong, J. Li, H. Wang, R. Liu, D. Liu, R. Zhou, T. Xie, R. Fu, and X. Duan, “A multitask model for realtime fish detection and segmentation based on YOLOv5,” PeerJ Computer Science , vol. 9, p. e1262, 2023

  10. [17]

    Fishtrack23: An ensemble underwater dataset for multi- object tracking,

    M. Dawkins, J. Prior, B. Lewis, R. Faillettaz, T. Banez, M. Salvi, A. Rollo, J. Simon, M. Campbell, M. Lucero, A. Chaudhary, B. Richards, and A. Hoogs, “Fishtrack23: An ensemble underwater dataset for multi- object tracking,” in 2024 IEEE/CVF Winter Conference on Applications ...

  11. [18]

    Ultralytics yolov8,

    G. Jocher, A. Chaurasia, and J. Qiu, “Ultralytics yolov8,” 2023. [Online]. Available: https://github.com/ultralytics/ultralytics 11 ABUJABAL et al. : FISHDET-M: A UNIFIED LARGE-SCALE BENCHMARK FOR ROBUST FISH DETECTION

  12. [19]

    Cascade r-cnn: High quality object detection and instance segmentation,

    Z. Cai and N. Vasconcelos, “Cascade r-cnn: High quality object detection and instance segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, p. 1–1, 2019

  13. [20]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in Computer Vision – ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part I . Berlin, Heidelberg: Springer-Verlag, 2...

  14. [21]

    Automated fish detection in underwater environments: Performance analysis of yolov8 and yolo- nas,

    K. Chieza, D. Brown, J. Connan, and D. Salie, “Automated fish detection in underwater environments: Performance analysis of yolov8 and yolo- nas,” in Artificial Intelligence Research , A. Gerber, J. Maritz, and A. W. Pillay, Eds. Cham: Springer Nature Switzerland, 2025, pp. 334–351

  15. [22]

    Fish population estimation and species classification from underwater video sequences using blob counting and shape analysis,

    J. Fabic, I. Turla, J. Capacillo, L. David, and P. C. Naval, “Fish population estimation and species classification from underwater video sequences using blob counting and shape analysis,” in 2013 IEEE International Underwater Technology Symposium (UT) , March 2013, pp. 1–6

  16. [23]

    Real-time and resource-efficient multi-scale adaptive robotics vision for underwater object detection and domain generalization,

    L. Saad Saoud, Z. Niu, L. Seneviratne, and I. Hussain, “Real-time and resource-efficient multi-scale adaptive robotics vision for underwater object detection and domain generalization,” in Proceedings of the IEEE International Conference on Image Processing (ICIP) , 2024, pp. ...

  17. [24]

    Enhanced fish species detection and classification using a novel deep learning approach,

    M. Iqtait, M. H. Alqaryouti, A. E. Sadeq, A. Aburomman, M. Baniata, Z. Mustafa, and H. Y . Chan, “Enhanced fish species detection and classification using a novel deep learning approach,”International Journal of Advanced Computer Science and Applications , vol. 15, no. 10, 2024

  18. [25]

    Overview of the lifeclef 2014 fish task,

    C. Spampinato, S. Palazzo, B. Boom, and R. B. Fisher, “Overview of the lifeclef 2014 fish task,” in Working Notes for CLEF 2014 Conference, Sheffield, UK, September 15-18, 2014 , ser. CEUR Workshop Proceedings, L. Cappellato, N. Ferro, M. Halvey, and W. Kraaij, Eds., vol. 1180...

  19. [26]

    Underwater detection: A brief survey and a new multitask dataset,

    Y . Wei, Y . Wang, B. Zhu, C. Lin, D. Wu, X. Xue, and R. Wang, “Underwater detection: A brief survey and a new multitask dataset,” International Journal of Network Dynamics and Intelligence , vol. 3, no. 4, p. 100025, 2024, published: 25 December 2024

  20. [27]

    Take good care of your fish: fish re-identification with synchronized multi-view camera system,

    S. Fan, C. Song, H. Feng, and Z. Yu, “Take good care of your fish: fish re-identification with synchronized multi-view camera system,” Frontiers in Marine Science , vol. V olume 11 - 2024, 2024

  21. [28]

    A realistic fish-habitat dataset to evaluate algorithms for underwater visual analysis,

    A. Saleh, I. H. Laradji, D. A. Konovalov, M. Bradley, D. Vazquez, and M. Sheaves, “A realistic fish-habitat dataset to evaluate algorithms for underwater visual analysis,” Scientific Reports, vol. 10, no. 1, p. 14671, 9 2020

  22. [29]

    Ozfish dataset - machine learning dataset for baited remote underwater video stations,

    Australian Institute of Marine Science (AIMS), University of Western Australia (UWA), and Curtin University, “Ozfish dataset - machine learning dataset for baited remote underwater video stations,” 2019, accessed 22-Jul-2025

  23. [30]

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

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster r-cnn: towards real-time object detection with region proposal networks,” p. 91–99, 2015

  24. [31]

    Seeing through the haze: A comprehensive review of underwater image enhancement techniques,

    L. Saad Saoud, M. Elmezain, A. Sultan, M. Heshmat, L. Seneviratne, and I. Hussain, “Seeing through the haze: A comprehensive review of underwater image enhancement techniques,” IEEE Access, vol. 12, pp. 145 206–145 233, 2024

  25. [32]

    Brackishmot: The brackish multi-object tracking dataset,

    M. Pedersen, D. Lehotský, I. Nikolov, and T. B. Moeslund, “Brackishmot: The brackish multi-object tracking dataset,” in Image Analysis, R. Gade, M. Felsberg, and J.-K. Kämäräinen, Eds. Cham: Springer Nature Switzerland, 2023, pp. 17–33

  26. [33]

    Trashcan: A semantically- segmented dataset towards visual detection of marine debris,

    J. Hong, M. Fulton, and J. Sattar, “Trashcan: A semantically- segmented dataset towards visual detection of marine debris,” arXiv preprint arXiv:2007.08097 , 2020. [Online]. Available: https: //arxiv.org/abs/2007.08097

  27. [34]

    Watermask: Instance segmentation for underwater imagery,

    S. Lian, H. Li, R. Cong, S. Li, W. Zhang, and S. Kwong, “Watermask: Instance segmentation for underwater imagery,” in 2023 IEEE/CVF International Conference on Computer Vision (ICCV) , Oct 2023, pp. 1305–1315

  28. [35]

    Eornet: An improved rotating box detection model for counting juvenile fish under occlusion and overlap,

    P. Zhang, L. Wang, G. Wang, and D. Li, “Eornet: An improved rotating box detection model for counting juvenile fish under occlusion and overlap,” Engineering Applications of Artificial Intelligence , vol. 124, p. 106528, 2023

  29. [36]

    Research on realizing the 3d occlusion tracking location method of fish’s school target,

    J. Mao, G. Xiao, W. Sheng, Z. Qu, and Y . Liu, “Research on realizing the 3d occlusion tracking location method of fish’s school target,” Neurocomputing, vol. 214, pp. 61–79, 2016

  30. [37]

    Fish classification using deep learning on small scale and low-quality images,

    M. Sudhakara, M. J. Meena, K. R. Madhavi, P. Anjaiah, and L. P. K, “Fish classification using deep learning on small scale and low-quality images,” International Journal of Intelligent Systems and Applications in Engineering , vol. 10, no. 1s, pp. 279–288, October 2022, resear...

  31. [38]

    Improving transfer learning and squeeze- and-excitation networks for small-scale fine-grained fish image classification,

    C. Qiu, S. Zhang, C. Wang, Z. Yu, H. Zheng, and B. Zheng, “Improving transfer learning and squeeze- and-excitation networks for small-scale fine-grained fish image classification,” IEEE Access, vol. 6, pp. 78 503– 78 512, 2018

  32. [39]

    Automatic discard registration in cluttered environments using deep learning and object tracking: class imbalance, occlusion, and a comparison to human review,

    R. van Essen, A. Mencarelli, A. van Helmond, L. Nguyen, J. Batsleer, J.-J. Poos, and G. Kootstra, “Automatic discard registration in cluttered environments using deep learning and object tracking: class imbalance, occlusion, and a comparison to human review,” ICES Journal of M...

  33. [40]

    Fine-grained fish classification from small to large datasets with vision transformers,

    R. J. M. Veiga and J. M. F. Rodrigues, “Fine-grained fish classification from small to large datasets with vision transformers,” IEEE Access , vol. 12, pp. 113 642–113 660, 2024

  34. [41]

    A dual-branch feature fusion neural network for fish image fine-grained recognition,

    X. Geng, J. Gao, Y . Zhang, and R. Wang, “A dual-branch feature fusion neural network for fish image fine-grained recognition,” The Visual Computer, vol. 40, no. 10, pp. 6883–6896, October 2024

  35. [42]

    Open-ocean fish reveal an omnidirectional solution to camouflage in polarized environments,

    P. C. Brady, A. A. Gilerson, G. W. Kattawar, J. M. Sullivan, M. S. Twar- dowski, H. M. Dierssen, M. Gao, K. Travis, R. I. Etheredge, A. Tonizzo, A. Ibrahim, C. Carrizo, Y . Gu, B. J. Russell, K. Mislinski, S. Zhao, and M. E. Cummings, “Open-ocean fish reveal an omnidirectional...

  36. [43]

    Active detection for fish species recognition in underwater environments,

    C. Shah, M. M. Nabi, S. Y . Alaba, R. Caillouet, J. Prior, M. Campbell, M. D. Grossi, F. Wallace, J. E. Ball, and R. Moorhead, “Active detection for fish species recognition in underwater environments,” in Ocean Sensing and Monitoring XVI , W. Hou and L. J. Mullen, Eds., vol. ...

  37. [44]

    Enhanced fish species identification using transfer learning on balanced datasets,

    S. Berlia, V . K. Singh, M. Kumar, R. Mahato, and M. Mishra, “Enhanced fish species identification using transfer learning on balanced datasets,” in 2024 15th International Conference on Computing Communication and Networking Technologies (ICCCNT) , June 2024, pp. 1–5

  38. [45]

    Few-shot fine-grained fish species classification via sandwich attention covamnet,

    J. Zhai, L. Han, Y . Xiao, M. Yan, Y . Wang, and X. Wang, “Few-shot fine-grained fish species classification via sandwich attention covamnet,” Frontiers in Marine Science , vol. V olume 10 - 2023, 2023

  39. [46]

    Comparative analysis of neural architectures for underwater object detection,

    S. T. K R, K. S. Ananda Kumar, S. P. R, and V . L, “Comparative analysis of neural architectures for underwater object detection,” in 2024 Second International Conference on Advances in Information Technology (ICAIT), vol. 1, July 2024, pp. 1–7

  40. [47]

    Enhanced detection and classification of underwater objects using rov and computer vision,

    M. Abdalhafez, I. M. H. AbdelDaiam, M. E. H. Eltaib, and M. Ab- delrahim, “Enhanced detection and classification of underwater objects using rov and computer vision,” JES. Journal of Engineering Sciences , vol. 52, no. 2, pp. 73–86, 2024

  41. [48]

    Underwater image quality evaluation: A comprehensive review,

    M. Shen, M. Yang, J. Zhong, H. Liu, and C. Pan, “Underwater image quality evaluation: A comprehensive review,” IET Image Processing , vol. 19, no. 1, p. e70068, 2025

  42. [49]

    Weighted multi-error information entropy based you only look once network for underwater object detection,

    H. Ma, Y . Zhang, S. Sun, W. Zhang, M. Fei, and H. Zhou, “Weighted multi-error information entropy based you only look once network for underwater object detection,” Engineering Applications of Artificial Intelligence, vol. 130, p. 107766, 2024

  43. [50]

    A novel underwater marine dataset with diverse scenarios for robust object detection,

    S. Bhalla, A. Kumar, and R. Kushwaha, “A novel underwater marine dataset with diverse scenarios for robust object detection,” in Proceedings of the 2024 Sixteenth International Conference on Contemporary Computing, ser. IC3-2024. New York, NY , USA: Association for Computing M...

  44. [51]

    A performance evaluation method for distant early warning sonar mobile area search,

    W. Liu, Q. Ma, P. Liu, and H. Zhao, “A performance evaluation method for distant early warning sonar mobile area search,” in 2023 3rd International Conference on Electronic Information Engineering and Computer Science (EIECS) , Sep. 2023, pp. 360–364

  45. [52]

    An underwater image quality assessment metric,

    P. Guo, H. Liu, D. Zeng, T. Xiang, L. Li, and K. Gu, “An underwater image quality assessment metric,” IEEE Transactions on Multimedia , vol. 25, pp. 5093–5106, 2023

  46. [53]

    Toward dimension- enriched underwater image quality assessment,

    Q. Jiang, X. Yi, L. Ouyang, J. Zhou, and Z. Wang, “Toward dimension- enriched underwater image quality assessment,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 35, no. 2, pp. 1385–1398, Feb 2025

  47. [54]

    Advancing underwater vision: A survey of deep learning models for underwater object recognition and tracking,

    M. Elmezain, L. Saad Saoud, A. Sultan, M. Heshmat, L. Seneviratne, and I. Hussain, “Advancing underwater vision: A survey of deep learning models for underwater object recognition and tracking,” IEEE Access, vol. 12, 2025, early Access

  48. [55]

    A visual servoing scheme for autonomous aquaculture net pens inspection using rov,

    W. Akram, A. Casavola, N. Kapetanovi ´c, and N. Miškovic, “A visual servoing scheme for autonomous aquaculture net pens inspection using rov,” Sensors, vol. 22, no. 9, p. 3525, 2022

  49. [56]

    Aquaculture defects recognition via multi-scale semantic segmentation,

    W. Akram, T. Hassan, H. Toubar, M. Ahmed, N. Miškovic, L. Seneviratne, and I. Hussain, “Aquaculture defects recognition via multi-scale semantic segmentation,” Expert systems with applications , vol. 237, p. 121197, 2024

  50. [57]

    Aquayolo: Advanced yolo-based 12 ABUJABAL et al. : FISHDET-M: A UNIFIED LARGE-SCALE BENCHMARK FOR ROBUST FISH DETECTION fish detection for optimized aquaculture pond monitoring,

    M. Vijayalakshmi and A. Sasithradevi, “Aquayolo: Advanced yolo-based 12 ABUJABAL et al. : FISHDET-M: A UNIFIED LARGE-SCALE BENCHMARK FOR ROBUST FISH DETECTION fish detection for optimized aquaculture pond monitoring,” Scientific Reports, vol. 15, no. 1, p. 6151, February 2025

  51. [58]

    Research on improved lightweight fish detection algorithm based on yolov8n,

    Q. Zhang and S. Chen, “Research on improved lightweight fish detection algorithm based on yolov8n,” Journal of Marine Science and Engineering, vol. 12, no. 10, 2024

  52. [59]

    Feedfirst: Intelligent monitoring system for indoor aquaculture tanks,

    B. Teixeira, A. P. Lima, C. Pinho, D. Viegas, N. Dias, H. Silva, and J. Almeida, “Feedfirst: Intelligent monitoring system for indoor aquaculture tanks,” in OCEANS 2022, Hampton Roads , Oct 2022, pp. 1–7

  53. [60]

    A school of robotic fish for mariculture monitoring in the sea coast,

    Y .-S. Ryuh, G.-H. Yang, J. Liu, and H. Hu, “A school of robotic fish for mariculture monitoring in the sea coast,” Journal of Bionic Engineering , vol. 12, no. 1, pp. 37–46, 2015

  54. [61]

    Research on robotic fish swarm network technology based on underwater acoustic communication,

    H. Wang, Z. Gao, B. Li, and N. Gao, “Research on robotic fish swarm network technology based on underwater acoustic communication,” in 2023 IEEE International Conference on Image Processing and Computer Applications (ICIPCA), Aug 2023, pp. 475–479

  55. [62]

    Vision-based autonomous navigation for unmanned surface vessel in extreme marine conditions,

    M. Ahmed, A. B. Bakht, T. Hassan, W. Akram, A. Humais, L. Seneviratne, S. He, D. Lin, and I. Hussain, “Vision-based autonomous navigation for unmanned surface vessel in extreme marine conditions,” in 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IRO...

  56. [63]

    Marine X: Design and implementation of unmanned surface vessel for vision guided navigation,

    M. U. Din, A. Humais, W. Akram, M. Alblooshi, L. Saad Saoud, A. Alblooshi, L. Seneviratne, and I. Hussain, “Marine X: Design and implementation of unmanned surface vessel for vision guided navigation,” in 2023 21st International Conference on Advanced Robotics (ICAR) . IEEE, 2...

  57. [64]

    Enhancing aquaculture net pen inspection: A benchmark study on detection and semantic segmentation,

    W. Akram, A. B. Bakht, M. U. Din, L. Seneviratne, and I. Hussain, “Enhancing aquaculture net pen inspection: A benchmark study on detection and semantic segmentation,” IEEE Access, 2024

  58. [65]

    Recognition and calculation of fish rafts in mariculture on the basis of artificial intelligence,

    W. Zhang, L. Zhang, Y . Zhong, P. Lin, and F. Zhang, “Recognition and calculation of fish rafts in mariculture on the basis of artificial intelligence,” in Proceedings of the 2023 International Conference on Wireless Communications, Networking and Applications , P. Siarry, M. ...

  59. [66]

    Marine aquaculture sites have huge potential as data providers for climate change assessments,

    L. Falconer, S. Halstensen, S. F. Rinø, C. Noble, T. Dale, R. Alvestad, and E. Ytteborg, “Marine aquaculture sites have huge potential as data providers for climate change assessments,” Aquaculture, vol. 595, p. 741519, 2025

  60. [67]

    Beyond observation: Deep learning for animal behavior and ecological conservation,

    L. Saad Saoud, A. Sultan, M. Elmezain, M. Heshmat, L. Seneviratne, and I. Hussain, “Beyond observation: Deep learning for animal behavior and ecological conservation,” Ecological Informatics, vol. 85, p. 102893, 2024

  61. [68]

    Aasnet: A novel image instance segmentation framework for fine-grained fish recognition via linear correlation attention and dynamic adaptive focal loss,

    J. Kong, S. Tang, J. Feng, L. Mo, and X. Jin, “Aasnet: A novel image instance segmentation framework for fine-grained fish recognition via linear correlation attention and dynamic adaptive focal loss,” Applied Sciences, vol. 15, no. 7, 2025

  62. [69]

    High-accuracy real-time fish detection based on self-build dataset and rird-yolov3,

    W. Wang, B. He, and L. Zhang, “High-accuracy real-time fish detection based on self-build dataset and rird-yolov3,” Complexity, vol. 2021, no. 1, p. 4761670, 2021

  63. [70]

    Benchmarking vision-based object tracking for usvs in complex maritime environments,

    M. U. Din, A. B. Bakht, W. Akram, Y . Dong, L. Seneviratne, and I. Hussain, “Benchmarking vision-based object tracking for usvs in complex maritime environments,” IEEE Access, 2025

  64. [71]

    Usod10k: A new benchmark dataset for underwater salient object detection,

    L. Hong, X. Wang, G. Zhang, and M. Zhao, “Usod10k: A new benchmark dataset for underwater salient object detection,” IEEE Transactions on Image Processing, vol. 34, pp. 1602–1615, 2025

  65. [72]

    Coco - common objects in context,

    COCO Consortium, “Coco - common objects in context,” 2014, accessed: 2025-07-15. [Online]. Available: https://cocodataset.org

  66. [73]

    Fishnet: A large-scale dataset and benchmark for fish recognition, detection, and functional trait prediction,

    F. F. Khan, X. Li, A. J. Temple, and M. Elhoseiny, “Fishnet: A large-scale dataset and benchmark for fish recognition, detection, and functional trait prediction,” in 2023 IEEE/CVF International Conference on Computer Vision (ICCV), 2023, pp. 20 439–20 449

  67. [74]

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

    A. Wang, H. Chen, L. Liu, Y . Wang, Y . Zhong, C. Shan, Z. Guo, C. Xu, and S. Chen, “Yolov10: Real-time end-to-end object detection,” arXiv preprint arXiv:2405.14458, 2024, https://arxiv.org/abs/2405.14458

  68. [75]

    Ultralytics yolo11,

    G. Jocher and J. Qiu, “Ultralytics yolo11,” 2024. [Online]. Available: https://github.com/ultralytics/ultralytics

  69. [76]

    Yolov12: Attention-centric real-time object detectors,

    Y . Tian, Q. Ye, and D. Doermann, “Yolov12: Attention-centric real-time object detectors,” 2025. [Online]. Available: https://github.com/ sunsmarterjie/yolov12

  70. [77]

    Super-gradients,

    S. Aharon, Louis-Dupont, Ofri Masad, K. Yurkova, Lotem Fridman, Lkdci, E. Khvedchenya, R. Rubin, N. Bagrov, B. Tymchenko, T. Keren, A. Zhilko, and Eran-Deci, “Super-gradients,” 2021

  71. [78]

    Sparse r-cnn: An end-to-end framework for object detection,

    P. Sun, R. Zhang, Y . Jiang, T. Kong, C. Xu, W. Zhan, M. Tomizuka, Z. Yuan, and P. Luo, “Sparse r-cnn: An end-to-end framework for object detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 12, pp. 15 650–15 664, Dec 2023

  72. [79]

    Detrs beat yolos on real-time object detection,

    Y . Zhao, W. Lv, S. Xu, J. Wei, G. Wang, Q. Dang, Y . Liu, and J. Chen, “Detrs beat yolos on real-time object detection,” pp. 16 965–16 974, June 2024

  73. [80]

    Deformable detr: Deformable transformers for end-to-end object detection,

    X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable detr: Deformable transformers for end-to-end object detection,” 2021. [Online]. Available: https://arxiv.org/abs/2010.04159

  74. [81]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Dollár, “Focal loss for dense object detection,” vol. 42, no. 2, Feb 2020, pp. 318–327

  75. [82]

    Ssd: Single shot multibox detector,

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y . Fu, and A. C. Berg, “Ssd: Single shot multibox detector,” pp. 21–37, 2016

  76. [83]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Köpf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, PyTorch: an imperative style, high- pe...

  77. [84]

    Ultralytics YOLO,

    G. Jocher, J. Qiu, and A. Chaurasia, “Ultralytics YOLO,” Jan. 2023. [Online]. Available: https://github.com/ultralytics/ultralytics

  78. [85]

    Mmdetection: Open mmlab detection toolbox and benchmark,

    K. Chen, J. Wang, J. Pang, Y . Cao, Y . Xiong, X. Li, S. Sun, W. Feng, Z. Liu, J. Xu, Z. Zhang, D. Cheng, C. Zhu, T. Cheng, Q. Zhao, B. Li, X. Lu, R. Zhu, Y . Wu, J. Dai, J. Wang, J. Shi, W. Ouyang, C. C. Loy, and D. Lin, “Mmdetection: Open mmlab detection toolbox and benchmar...

  79. [86]

    pycocotools: Coco api for python,

    T.-Y . Lin and M. C. Consortium, “pycocotools: Coco api for python,” https://github.com/cocodataset/cocoapi, 2015, accessed: 2025-07-15

  80. [87]

    Fish4knowledge dataset dataset,

    g18L5754, “Fish4knowledge dataset dataset,” https://universe.roboflow. com/g18l5754/fish4knowledge-dataset, oct 2023, visited on 2025- 05-21. [Online]. Available: https://universe.roboflow.com/g18l5754/ fish4knowledge-dataset

  81. [88]

    Fish video dataset,

    A. MURME, “Fish video dataset,” https://universe.roboflow.com/ aarjoo-murme/fish-video-ls42k, nov 2023, visited on 2025-05-

  82. [89]

    Available: https://universe.roboflow.com/aarjoo-murme/ fish-video-ls42k

    [Online]. Available: https://universe.roboflow.com/aarjoo-murme/ fish-video-ls42k

  83. [90]

    Fish-video dataset,

    seoultech, “Fish-video dataset,” https://universe.roboflow.com/seoultech/ fish-video, may 2022, visited on 2025-05-21. [Online]. Available: https://universe.roboflow.com/seoultech/fish-video

  84. [91]

    Automatic detection and classification of coastal mediterranean fish from underwater images: Good practices for robust training,

    I. A. Catalán, A. Álvarez Ellacuría, J.-L. Lisani, J. Sánchez, G. Vizoso, A. E. Heinrichs-Maquilón, H. Hinz, J. Alós, M. Signarioli, J. Aguzzi, M. Francescangeli, and M. Palmer, “Automatic detection and classification of coastal mediterranean fish from underwater images: Good ...

  85. [92]

    detect aqurium dataset,

    aquarium, “detect aqurium dataset,” https://universe.roboflow.com/ aquarium-lrui2/detect-aqurium, sep 2023, visited on 2025-06-30. [Online]. Available: https://universe.roboflow.com/aquarium-lrui2/detect-aqurium

  86. [93]

    Aquatic animal species (aas),

    M.-Q. Le, T.-N. Le, T. V . Nguyen, I. Echizen, and M.-T. Tran, “Aquatic animal species (aas),” 2023, data set

  87. [94]

    Fish clean dataset,

    PAFD, “Fish clean dataset,” https://universe.roboflow.com/pafd/fish-clean, feb 2023, visited on 2025-06-30. [Online]. Available: https://universe. roboflow.com/pafd/fish-clean

  88. [96]

    fish_dataset_florence_1 dataset,

    test, “fish_dataset_florence_1 dataset,” https://universe.roboflow.com/ test-fflhx/fish_dataset_florence_1, Aug. 2024, accessed on 2025-07-20

  89. [97]

    Eba-ai: Ethics-guided bias-aware ai for efficient underwater image enhancement and coral reef monitoring,

    L. Saad Saoud and I. Hussain, “Eba-ai: Ethics-guided bias-aware ai for efficient underwater image enhancement and coral reef monitoring,”

  90. [98]

    Deepfins: Capturing dynamics in underwater videos for fish detection,

    A. Jalal, A. Salman, A. Mian, S. Ghafoor, and F. Shafait, “Deepfins: Capturing dynamics in underwater videos for fish detection,” Ecological Informatics, vol. 86, p. 103013, 2025. 13

  91. [99]

    Automatic fish detection in underwater videos by a deep neural network-based hybrid motion learning system,

    A. Salman, S. A. Siddiqui, F. Shafait, A. Mian, M. R. Shortis, K. Khurshid, A. Ulges, and U. Schwanecke, “Automatic fish detection in underwater videos by a deep neural network-based hybrid motion learning system,” ICES Journal of Marine Science , vol. 77, no. 4, pp. 1295–1307...

  92. [2025]

    Available: https://arxiv.org/abs/2507.15036

    [Online]. Available: https://arxiv.org/abs/2507.15036

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.