Pith. sign in

REVIEW 4 major objections 6 minor 42 references

USIS16K: High-Quality Dataset for Underwater Salient Instance Segmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read USIS16K provides 16,151 underwater images with instance-level salient object masks covering 158 categories.

desk verdict USIS16K is a genuinely new, potentially useful 158-category underwater dataset, but the paper's central 'high-quality' claim rests on an annotation pipeline with zero measured quality control; worth reviewing, not worth believing yet. read the letter →

arxiv 2506.19472 v2 pith:3ZIGM3MY submitted 2025-06-24 cs.CV

Lin Hong , Xin Wang , Yihao Li , Xia Wang This is my paper
classification cs.CV
keywords underwatersalientinstancesegmentationobjectdetectioneye-trackingannotationdatasetbenchmarkmulti-levelimagery
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 introduces USIS16K, a large-scale dataset for underwater salient instance segmentation (USIS), built from more than 35,000 candidate images that were filtered down to 16,151 high-resolution underwater images. The images span 158 object categories grouped into 11 super-categories, and each image carries multi-level annotations: classification, gaze points, bounding boxes, pixel-wise salient object masks, and instance-level masks. The authors' central claim is that this combination of category diversity, scene complexity, and aligned labels makes USIS16K the most complete public resource for jointly studying where to look and what is there in underwater images. They also report benchmark results for eight object detectors and twelve instance segmentation methods, establishing baselines for future USIS models. A sympathetic reader would care because underwater vision lacks the large, densely annotated resources that drove progress on land.

What carries the argument

The load-bearing object is the dataset itself, constructed through four steps: internet and in-situ image collection, volunteer filtering, ten-annotator multi-level annotation, and category-level splitting. The distinctive mechanism is the annotation pipeline: eye-tracking data first identify gaze points, these guide the determination of salient objects, and then classification, bounding boxes, salient object segmentation, and instance-level masks are layered in sequence, producing aligned ground truth for detection, segmentation, and saliency. A hierarchical taxonomy of 11 super-categories and 158 classes organizes the label space, and benchmark evaluations of eight detectors and twelve segmentation models supply the reference numbers.

What would settle it

Compute inter-annotator agreement on a random sample by having fresh annotators mark salient objects without eye-tracking guidance, or compare the gaze-based saliency decisions against an independent eye-tracking study on the same images; if agreement is near chance or gaze predictions do not match independent saliency judgments, the claim that the masks are high-quality salient instance annotations would not be supported.

Watch

Extended reading notes

Core claim

The paper's core claim is that USIS16K fills the gap in underwater vision by supplying the largest category-diverse set of salient instance segmentation annotations: 16,151 images, 158 categories, and a category-level 7:2:1 split into 11,306 training, 3,230 validation, and 1,615 test images. Annotations were produced by ten trained annotators in a coarse-to-fine pipeline, with eye-tracking data used to decide which objects are salient. Object sizes span [0.05%, 93.98%] of image area, images range up to 7500 by 5184 pixels, and the red channel is consistently the lowest, reflecting underwater light attenuation. On this dataset, the paper reports that Cascade R-CNN leads the detection benchmark across mAP, AP75, APM, and APL, while USIS-SAM, ConvNeXt, and PointRend lead instance segmentation mAP; USIS-SAM and WaterMask, as domain-specific models, outperform most methods developed for terrestrial environments.

Load-bearing premise

The load-bearing premise is that eye-tracking data, applied by ten trained annotators, are a valid and consistent signal for which objects are genuinely salient in underwater images; the paper reports no independent validation or inter-annotator agreement for this guidance.

Editorial extensions

If this is right

  • Underwater salient instance segmentation methods can be trained and compared on a common set of 16,151 images instead of smaller, category-limited sets.
  • The aligned multi-level labels allow one dataset to support object detection, salient object detection, instance segmentation, and combined USIS evaluation.
  • The 158-category taxonomy and broad resolution and background diversity give models a harder generalization test than prior underwater datasets.
  • The reported detection and segmentation benchmarks provide reference numbers that future USIS models can be measured against.
  • Category-level 7:2:1 splitting reduces the risk of category imbalance distorting evaluation.

Reading between the lines

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

  • If gaze-guided saliency proves reliable, the dataset could be used to study how human attention maps onto segmentation masks, a link most segmentation datasets lack.
  • Because images span up to 7500 pixels in width, USIS16K may also serve as a stress test for resolution-robust detectors and segmentation heads.
  • The benchmark gap between terrestrial and domain-specific models suggests room for methods that jointly model waterbody color shifts and saliency; this is an editorial extension, not a claim the paper makes.
  • A natural testable extension is to release per-image gaze maps alongside masks, letting later work verify or refine the saliency ground truth.
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

4 major / 6 minor

Summary. The paper introduces USIS16K, a dataset of 16,151 underwater images with multi-level annotations (eye-tracking gaze points, classification, bounding boxes, salient object masks, and instance-level masks) spanning 158 object categories, and reports benchmark evaluations for 8 object detectors and 12 instance segmentation methods. The split statistics are internally consistent (11,306/3,230/1,615), as are the object-size and object-count aggregates. The central claim is that the annotations are 'high-quality', but this is supported only by a qualitative description of the annotation pipeline and is not backed by quantitative quality metrics.

Significance. If the annotation-quality claim can be substantiated, USIS16K would be a valuable resource for underwater salient instance segmentation: it is the most category-diverse underwater dataset among those compared in Table 1, it provides aligned labels for detection, segmentation, and saliency, and the paper contributes a broad benchmark of 20 models with public release of the data. The aggregate statistics are internally consistent, and the benchmark numbers are plausible. However, the paper's central claim of 'high-quality' annotations is currently asserted rather than demonstrated, and the provenance of a subset of the images is not clearly disclosed. These issues need to be resolved before the dataset can be relied upon as a benchmark.

major comments (4)
  1. [Abstract; Section 2.1.3] The central claim that USIS16K contains 'high-quality' instance-level salient object masks is never measured. Section 2.1.3 describes a team of ten trained annotators guided by eye-tracking data, but reports no inter-annotator agreement (IoU, Cohen's kappa, or boundary error), no comparison against a held-out expert consensus, no eye-tracking protocol details (number of participants, device, gaze-to-saliency mapping), and no annotation QA statistics. Since every benchmark result in Section 3 is computed against these masks, unvalidated annotation quality would transfer directly into the reported scores. Please add quantitative annotation-quality evidence or explicitly qualify the claim.
  2. [Section 2.1.1] The provenance of a subset of images is unclear: the text states that 'representative underwater images from the USIS16K dataset [12] are incorporated', but reference [12] is USOD10K, not USIS16K. This appears to be a misnaming, and the paper does not disclose what fraction of USIS16K images come from USOD10K, whether those images were re-annotated, or whether they were used in prior benchmarks in a way that could affect train/test separation. Please clarify the source and quantify any overlap with existing datasets.
  3. [Section 2.2.1; Table 2] The super-category description is internally inconsistent. The text states that 11 super-categories were defined and then lists eight names ('obstacles, facilities, underwater animals, humans, relics, marine fishes, plants, and litter'), while Table 2 lists eleven different names ('Human participant', 'Fish', 'Shellfish', 'Marine animals', 'Underwater trash', 'Underwater vehicles', 'Lost items', 'Underwater facilities', 'Underwater relics', 'Underwater wrecks', 'Others'). The hierarchical taxonomy in Figure 3 should be aligned with Table 2 and ideally released in a machine-readable form.
  4. [Section 3; Tables 4 and 5] The benchmark section is not reproducible as written. Table 4 assigns incorrect reference numbers to several methods (e.g., SSD is cited as [11], RetinaNet as [13], ATSS as [4], YOLOF as [2], Faster R-CNN as [36], Cascade R-CNN as [37], Libra R-CNN as [16], and Dynamic R-CNN as [32], none of which match the reference list). In addition, the paper gives no training details such as epochs, hardware, input resolution, augmentation, or evaluation protocol, so the reported baselines cannot be reproduced. Please correct the citations and add a benchmark protocol subsection.
minor comments (6)
  1. [Abstract; Section 2.3.5] The dataset is described as 'high-resolution', but the reported resolution range includes images as small as 225×135 pixels; please define the resolution criterion used.
  2. [Figure 7] The caption says '(e) Resolution distribution' but the panel is labeled (b); please correct the panel labels.
  3. [Section 3.2] The text refers to 'YOLOCAT' in the discussion of Figure 9; this appears to be a typo for YOLACT and should be corrected.
  4. [Table 5] Numerical entries such as '.736' and '.900' should be written as '0.736' and '0.900' for readability and consistency with Table 4.
  5. [References] References [8] and [9] are duplicates of the same paper ('Salient objects in clutter'); please merge or remove one.
  6. [Section 2.1.1] The phrase 'USIS16K dataset [12]' should read 'USOD10K dataset [12]' if that is the intended source, and the overlap should be stated explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: USIS16K is an empirical dataset contribution, and the reported benchmarks come from retraining standard models on the new data rather than from any fitted parameter or self-citation chain.

full rationale

The paper's central claims are dataset construction and benchmark evaluation. The derivation chain is empirical: images were collected from the internet, from the authors' prior USOD10K dataset, and from in-situ robotic captures; filtering was done by volunteers; annotation used eye-tracking guidance followed by bounding boxes, segmentation, and instance masks; and benchmarks were produced by retraining 8 detection and 12 instance segmentation methods with their default configurations on the USIS16K splits. No equation in the paper defines a predicted quantity in terms of a fitted parameter, and no benchmark result is shown to equal its training input by construction. The self-citation to USOD10K [12] is used only as an image source, not as a load-bearing theorem or as a justification for the correctness of the annotations, so it does not make the argument circular. The absence of inter-annotator agreement or eye-tracking validation is a legitimate concern about annotation quality, but it is a validity gap, not a circular-derivation issue. Accordingly, the circularity score is 0.

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

The central contribution is an empirical resource rather than a mathematical derivation, so there are no fitted parameters or invented physical entities. The ledger instead records the domain assumptions about annotation validity, representativeness, and split balance that the 'high-quality dataset' claim depends on.

assumptions (4)
  • domain assumption Eye-tracking data identify the true salient regions in underwater images.
    Section 2.1.3 states annotation begins with identifying salient objects guided by eye-tracking data, but no validation is provided for underwater scenes.
  • domain assumption Three volunteers and ten annotators, after brief training, produce consistent and accurate filtering and labeling.
    Sections 2.1.2 and 2.1.3 assume consistency and accuracy without inter-annotator agreement or QA statistics.
  • domain assumption Internet-sourced, robot-captured, and USOD10K-derived images are representative of real-world underwater environments and object categories.
    Section 2.1.1 describes collection without a sampling protocol or bias analysis.
  • domain assumption The 7:2:1 category-level split produces adequately balanced training, validation, and test sets.
    Section 2.1.4 asserts balance but does not report per-category counts across splits.

how reviews work

0 comments
Cite this review

Pith. "Pith review of USIS16K: High-Quality Dataset for Underwater Salient Instance Segmentation." pith.science (2026). https://pith.science/paper/3ZIGM3MY

@misc{pith2026250619472,
  author       = {Pith},
  title        = {Pith review of: USIS16K: High-Quality Dataset for Underwater Salient Instance Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3ZIGM3MY}},
  note         = {Machine review of arXiv:2506.19472}
}
read the original abstract

Inspired by the biological visual system that selectively allocates attention to efficiently identify salient objects or regions, underwater salient instance segmentation (USIS) aims to jointly address the problems of where to look (saliency prediction) and what is there (instance segmentation) in underwater scenarios. However, USIS remains an underexplored challenge due to the inaccessibility and dynamic nature of underwater environments, as well as the scarcity of large-scale, high-quality annotated datasets. In this paper, we introduce USIS16K, a large-scale dataset comprising 16,151 high-resolution underwater images collected from diverse environmental settings and covering 158 categories of underwater objects. Each image is annotated with high-quality instance-level salient object masks, representing a significant advance in terms of diversity, complexity, and scalability. Furthermore, we provide benchmark evaluations on underwater object detection and USIS tasks using USIS16K. To facilitate future research in this domain, the dataset and benchmark models are publicly available.

Figures

Figures reproduced from arXiv: 2506.19472 by the authors.

Figure 1
Figure 1. Underwater images in the USIS16K dataset are annotated with multi-level labels. From top to bottom: raw image with classifi [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Image annotation pipeline: (a) underwater image, (b) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Hierarchical taxonomic system of USIS16K dataset. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Number of underwater images for each salient object [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 7
Figure 7. Figure 7: Essential characteristics of the USIS16K dataset. (a) [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 8
Figure 8. Figure 8: PR curve comparison of 8 object detection methods eval [PITH_FULL_IMAGE:figures/full_fig_p005_8.png]
Figure 9
Figure 9. Figure 9: PR curve comparison of 12 instance segmentation [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 10
Figure 10. Figure 10: Qualitative comparison of 12 instance segmentation methods on the USIS16K dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 36 canonical work pages

  1. [12]

    Usod10k: A new benchmark dataset for underwater salient object de- tection

    Lin Hong, Xin Wang, Gan Zhang, and Ming Zhao. Usod10k: A new benchmark dataset for underwater salient object de- tection. IEEE Transactions on Image Processing, pages 1–1,

    work page

  2. [11]

    Mask r-cnn

    Kaiming He, Georgia Gkioxari, Piotr Doll ´ar, and Ross Gir- shick. Mask r-cnn. In Proceedings of the IEEE international conference on computer vision, pages 2961–2969, 2017. 1, 5, 6

    work page 2017

  3. [13]

    Mask scoring r-cnn

    Zhaojin Huang, Lichao Huang, Yongchao Gong, Chang Huang, and Xinggang Wang. Mask scoring r-cnn. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 6409–6418, 2019. 5, 6

    work page 2019

  4. [4]

    Hybrid task cascade for instance seg- mentation

    Kai Chen, Jiangmiao Pang, Jiaqi Wang, Yu Xiong, Xiaox- iao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jianping Shi, Wanli Ouyang, et al. Hybrid task cascade for instance seg- mentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4974–4983,

  5. [2]

    Yolact: Real-time instance segmentation

    Daniel Bolya, Chong Zhou, Fanyi Xiao, and Yong Jae Lee. Yolact: Real-time instance segmentation. In Proceedings of the IEEE/CVF international conference on computer vision, pages 9157–9166, 2019. 5, 6

    work page 2019

  6. [36]

    Solo: Segmenting objects by locations

    Xinlong Wang, Tao Kong, Chunhua Shen, Yuning Jiang, and Lei Li. Solo: Segmenting objects by locations. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVIII 16, pages 649–665. Springer, 2020. 5, 6

    work page 2020

  7. [37]

    Solov2: Dynamic and fast instance segmenta- tion

    Xinlong Wang, Rufeng Zhang, Tao Kong, Lei Li, and Chun- hua Shen. Solov2: Dynamic and fast instance segmenta- tion. Advances in Neural information processing systems , 33:17721–17732, 2020. 5, 6

    work page 2020

  8. [16]

    Pointrend: Image segmentation as rendering

    Alexander Kirillov, Yuxin Wu, Kaiming He, and Ross Gir- shick. Pointrend: Image segmentation as rendering. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9799–9808, 2020. 5, 6

    work page 2020

  9. [32]

    Detectors: Detecting objects with recursive feature pyramid and switch- able atrous convolution

    Siyuan Qiao, Liang-Chieh Chen, and Alan Yuille. Detectors: Detecting objects with recursive feature pyramid and switch- able atrous convolution. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 10213–10224, 2021. 5, 6

    work page 2021

Show all 42 references
  1. [1]

    A revised underwater image formation model

    Derya Akkaynak and Tali Treibitz. A revised underwater image formation model. In Proceedings of the IEEE con- ference on computer vision and pattern recognition , pages 6723–6732, 2018. 1

    2018

  2. [3]

    Cascade r-cnn: Delv- ing into high quality object detection

    Zhaowei Cai and Nuno Vasconcelos. Cascade r-cnn: Delv- ing into high quality object detection. In Proceedings of the Figure 10. Qualitative comparison of 12 instance segmentation methods on the USIS16K dataset. IEEE conference on computer vision and pattern recogni- tion, page...

  3. [5]

    You only look one-level feature

    Qiang Chen, Yingming Wang, Tong Yang, Xiangyu Zhang, Jian Cheng, and Jian Sun. You only look one-level feature. In IEEE Conference on Computer Vision and Pattern Recog- nition, 2021. 5

    2021

  4. [6]

    Sparse instance activation for real-time instance segmentation

    Tianheng Cheng, Xinggang Wang, Shaoyu Chen, Wenqiang Zhang, Qian Zhang, Chang Huang, Zhaoxiang Zhang, and Wenyu Liu. Sparse instance activation for real-time instance segmentation. In Proc. IEEE Conf. Computer Vision and Pat- tern Recognition (CVPR), 2022. 5, 6

    2022

  5. [7]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 2

    2009

  6. [8]

    Salient objects in clut- ter: Bringing salient object detection to the foreground

    Deng-Ping Fan, Ming-Ming Cheng, Jiang-Jiang Liu, Shang- Hua Gao, Qibin Hou, and Ali Borji. Salient objects in clut- ter: Bringing salient object detection to the foreground. In Proceedings of the European conference on computer vision (ECCV), pages 186–202, 2018. 1, 2

    2018

  7. [9]

    Salient objects in clutter: Bringing salient object detection to the foreground

    Deng-Ping Fan, Ming-Ming Cheng, Jiang-Jiang Liu, Shang- Hua Gao, Qibin Hou, and Ali Borji. Salient objects in clutter: Bringing salient object detection to the foreground. In Eu- ropean Conference on Computer Vision (ECCV) . Springer,

  8. [10]

    A survey on instance segmentation: state of the art.International jour- nal of multimedia information retrieval, 9(3):171–189, 2020

    Abdul Mueed Hafiz and Ghulam Mohiuddin Bhat. A survey on instance segmentation: state of the art.International jour- nal of multimedia information retrieval, 9(3):171–189, 2020. 4

    2020

  9. [14]

    Semantic segmentation of underwater im- agery: Dataset and benchmark

    Md Jahidul Islam, Chelsey Edge, Yuyang Xiao, Peigen Luo, Muntaqim Mehtaz, Christopher Morse, Sadman Sakib Enan, and Junaed Sattar. Semantic segmentation of underwater im- agery: Dataset and benchmark. In 2020 IEEE/RSJ Interna- tional Conference on Intelligent Robots and System...

    2020

  10. [15]

    SV AM: Saliency-guided Visual Attention Modeling by Autonomous Underwater Robot

    Md Jahidul Islam, Ruobing Wang, and Junaed Sattar. SV AM: Saliency-guided Visual Attention Modeling by Autonomous Underwater Robot. In Proceedings of Robotics: Science and Systems, New York City, NY , USA, June 2022. 2

    2022

  11. [17]

    Instance- level salient object segmentation

    Guanbin Li, Yuan Xie, Liang Lin, and Yizhou Yu. Instance- level salient object segmentation. In Proceedings of the IEEE conference on computer vision and pattern recogni- tion, pages 2386–2395, 2017. 1, 2

    2017

  12. [18]

    Uwsam: Segment anything model guided under- water instance segmentation and a large-scale benchmark dataset

    Hua Li, Shijie Lian, Zhiyuan Li, Runmin Cong, and Sam Kwong. Uwsam: Segment anything model guided under- water instance segmentation and a large-scale benchmark dataset. arXiv preprint arXiv:2505.15581, 2025. 2

    2025

  13. [19]

    Ion: Instance-level object navigation

    Weijie Li, Xinhang Song, Yubing Bai, Sixian Zhang, and Shuqiang Jiang. Ion: Instance-level object navigation. In Proceedings of the 29th ACM international conference on multimedia, pages 4343–4352, 2021. 1

    2021

  14. [20]

    Rehg, and Alan L

    Yin Li, Xiaodi Hou, Christof Koch, James M. Rehg, and Alan L. Yuille. The secrets of salient object segmentation. In 2014 IEEE Conference on Computer Vision and Pattern Recognition, pages 280–287, 2014. 4

    2014

  15. [21]

    Watermask: Instance segmentation for under- water imagery

    Shijie Lian, Hua Li, Runmin Cong, Suqi Li, Wei Zhang, and Sam Kwong. Watermask: Instance segmentation for under- water imagery. In Proceedings of the IEEE/CVF Interna- tional Conference on Computer Vision , pages 1305–1315,

  16. [22]

    Diving into underwater: Segment anything model guided underwater salient instance segmentation and a large-scale dataset.arXiv preprint arXiv:2406.06039, 2024

    Shijie Lian, Ziyi Zhang, Hua Li, Wenjie Li, Laurence Tian- ruo Yang, Sam Kwong, and Runmin Cong. Diving into underwater: Segment anything model guided underwater salient instance segmentation and a large-scale dataset.arXiv preprint arXiv:2406.06039, 2024. 1, 2, 5, 6

    2024 arXiv

  17. [23]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll´ar. Focal loss for dense object detection. In Pro- ceedings of the IEEE international conference on computer vision, 2017. 5

    2017

  18. [24]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision , pages 740–755. Springer, 2014. 2

    2014

  19. [25]

    Scg: Saliency and contour guided salient instance segmentation

    Nian Liu, Wangbo Zhao, Ling Shao, and Junwei Han. Scg: Saliency and contour guided salient instance segmentation. IEEE Transactions on Image Processing , 30:5862–5874,

  20. [26]

    Ssd: Single shot multibox detector

    Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, and Alexander C Berg. Ssd: Single shot multibox detector. In Computer Vision–ECCV 2016: 14th European Conference, Amster- dam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14,...

    2016

  21. [27]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feicht- enhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 11976–11986,

  22. [28]

    C. D. Mobley. Light and water: Radiative transfer in natural waters. Academic Press, 1994. 4

    1994

  23. [29]

    Libra r-cnn: Towards balanced learning for object detection

    Jiangmiao Pang, Kai Chen, Jianping Shi, Huajun Feng, Wanli Ouyang, and Dahua Lin. Libra r-cnn: Towards balanced learning for object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 821–830, 2019. 5

    2019

  24. [30]

    Transformer-based efficient salient instance segmentation networks with orientative query

    Jialun Pei, Tianyang Cheng, He Tang, and Chuanbo Chen. Transformer-based efficient salient instance segmentation networks with orientative query. IEEE Transactions on Mul- timedia, 25:1964–1978, 2022. 1

    1964

  25. [31]

    Calibnet: Dual- branch cross-modal calibration for rgb-d salient instance segmentation

    Jialun Pei, Tao Jiang, He Tang, Nian Liu, Yueming Jin, Deng-Ping Fan, and Pheng-Ann Heng. Calibnet: Dual- branch cross-modal calibration for rgb-d salient instance segmentation. IEEE Transactions on Image Processing , 33:4348–4362, 2024. 1, 2

    2024

  26. [33]

    You only look once: Unified, real-time object de- tection

    Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object de- tection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788, 2016. 1

    2016

  27. [34]

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

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, Jun 2017. 5

    2017

  28. [35]

    Scnet: Train- ing inference sample consistency for instance segmentation

    Thang Vu, Haeyong Kang, and Chang D Yoo. Scnet: Train- ing inference sample consistency for instance segmentation. In Proceedings of the AAAI Conference on Artificial Intelli- gence, volume 35, pages 2701–2709, 2021. 5, 6

    2021

  29. [38]

    Semantic segmen- tation in underwater ship inspections: Benchmark and data set

    Maryna Waszak, Alexandre Cardaillac, Brian Elvesæter, Frode Rødølen, and Martin Ludvigsen. Semantic segmen- tation in underwater ship inspections: Benchmark and data set. IEEE Journal of Oceanic Engineering , 48(2):462–473,

  30. [39]

    Unseen object instance segmentation for robotic en- vironments

    Christopher Xie, Yu Xiang, Arsalan Mousavian, and Dieter Fox. Unseen object instance segmentation for robotic en- vironments. IEEE Transactions on Robotics , 37(5):1343– 1359, 2021. 1

    2021

  31. [40]

    Dynamic r-cnn: Towards high quality object detection via dynamic training

    Hongkai Zhang, Hong Chang, Bingpeng Ma, Naiyan Wang, and Xilin Chen. Dynamic r-cnn: Towards high quality object detection via dynamic training. In Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23– 28, 2020, Proceedings, Part XV 16, pages 260–275. Springer,

    2020

  32. [41]

    Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection

    Shifeng Zhang, Cheng Chi, Yongqiang Yao, Zhen Lei, and Stan Z Li. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection. In Proceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 9759–9768, 2020. 5

    2020

  33. [42]

    Coralscop: Segment any coral image on this planet

    Ziqiang Zheng, Haixin Liang, Binh-Son Hua, Yue Him Wong, Put Ang, Apple Pui Yi Chui, and Sai-Kit Yeung. Coralscop: Segment any coral image on this planet. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 28170–28180, 2024. 1, 2

    2024

Pith tools

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