pith. sign in

arxiv: 2606.28405 · v1 · pith:NQFET3YInew · submitted 2026-06-24 · 💻 cs.CV

Enhancing Layer Interaction Using Key-Correlated Layer Attention

Jianlong Xiong , ChuanBo Xie , Le Yu , Quansong He , Tao He This is my paper

Pith reviewed 2026-06-30 01:07 UTC · model grok-4.3

classification 💻 cs.CV
keywords layer attentionkey correlationlinear complexityneural network architectureimage recognitionobject detectionmedical image segmentation
0
0 comments X

The pith

Key-Correlated Layer Attention reduces cross-layer interaction to linear complexity while retaining dynamic updates.

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

The paper proposes Key-Correlated Layer Attention to solve the quadratic cost of standard layer attention, where each layer queries all previous layers. It starts from the observation that key representations show high cosine similarity and uses this to derive a linear form directly from the original attention definition. This keeps dynamic information updates and long-range connections intact, unlike recurrent or other linear approximations. Spatial complexity stays fixed regardless of depth. Experiments on recognition, detection, and segmentation tasks show the method delivers competitive results.

Core claim

KCLA is obtained by replacing the full key-query interaction in layer attention with a correlated-key form, yielding linear time complexity, preserved dynamic updates, maintained long-range dependencies, and constant spatial cost independent of network depth.

What carries the argument

Key-Correlated Layer Attention (KCLA), which exploits high cosine similarity among key representations to linearize the attention computation while following the original layer attention definition.

If this is right

  • Deeper networks can add cross-layer connections without quadratic runtime increase.
  • Dynamic information flow between layers remains possible at linear cost.
  • Spatial memory usage does not grow with added layers.
  • The same mechanism supports image recognition, object detection, and medical segmentation without task-specific redesign.

Where Pith is reading between the lines

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

  • The linear form might transfer to transformer-style models where key similarity also appears.
  • A direct test on networks deeper than those in the paper would confirm the claimed depth-independent spatial cost.
  • If the similarity property holds in non-vision domains, the same derivation could apply to sequence models.

Load-bearing premise

Key representations across layers exhibit high cosine similarity.

What would settle it

Train a network, measure actual cosine similarity of its keys, and test whether KCLA performance collapses relative to full layer attention when similarity falls below the assumed high level.

Figures

Figures reproduced from arXiv: 2606.28405 by ChuanBo Xie, Jianlong Xiong, Le Yu, Quansong He, Tao He.

Figure 1
Figure 1. Figure 1: The details of the standard layer attention module (a) and RLA module [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of different layer attention methods. * marks methods [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The means and standard deviations of β for different layers in MRLA￾L, with error bars representing the corresponding standard deviations. The lower part of the error bars is truncated where it would exceed the figure boundary. Among them, (a) is tested at a depth of 56, and (b) is tested at a depth of 110. Quantization on K, which is employed in Transformer-VQ [11]. III. PROBLEM FORMULATION A. Review stan… view at source ↗
Figure 5
Figure 5. Figure 5: Implementation details of KCLA. Since f t V (·) is a linear transformation acting on all layer outputs. We factor f t V (·) out of the summation term. This is mathematically equivalent while retaining the dynamics of the standard layer attention, as the early layer outputs will also be transformed. When the Key representations of different layers are completely correlated, the approximation in Equation (7)… view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of the computational load and memory cost introduced [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The distribution of scores for different heads in KCLA-110. In this visualization, each row of small squares represents the attention distribution results, [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
read the original abstract

Recent advances in network architecture design have introduced layer attention to enhance inter-layer interactions. In such frameworks, each layer queries all preceding layers to establish cross-layer connections. However, layer attention results in quadratic computational complexity with respect to network depth. To mitigate this issue, prior works have proposed Recurrent Layer Attention (RLA) and linear attention mechanisms, which suffer from static information updates and limited long-range cross-layer dependency modeling. To overcome these limitations, we propose Key-Correlated Layer Attention (KCLA), inspired by our observation that Key representations in layer attention exhibit high cosine similarity. KCLA achieves linear computational complexity while preserving dynamic information updates, directly derived from the foundational definition of layer attention. Furthermore, KCLA maintains long-range cross-layer connections and features a fixed spatial complexity, independent of network depth. Empirical evaluations demonstrate that KCLA delivers good performance across diverse tasks, including image recognition, object detection, and medical image segmentation. The code is publicly available at https://github.com/bgx666/KCLA.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript proposes Key-Correlated Layer Attention (KCLA) to address the quadratic complexity of standard layer attention in deep networks. Based on the observation that key representations exhibit high cosine similarity across layers, the authors derive KCLA which achieves linear computational complexity, preserves dynamic information updates, maintains long-range cross-layer connections, and has fixed spatial complexity independent of depth. The method is evaluated on image recognition, object detection, and medical image segmentation tasks, with code made publicly available.

Significance. If the central derivation holds and the similarity assumption is robust, KCLA could offer an efficient alternative for modeling inter-layer dependencies without the drawbacks of recurrent or standard linear attentions. The public code availability is a positive for reproducibility. However, the significance depends on validating the load-bearing assumption about key similarity.

major comments (1)
  1. Abstract: The claim that KCLA is 'directly derived from the foundational definition of layer attention' and achieves linear complexity while preserving dynamic updates hinges on the high cosine similarity of keys. The abstract presents this similarity as an empirical observation without providing quantitative analysis, bounds, or proof that it holds independently of network depth or task, which is necessary to confirm the derivation is exact rather than approximate.

Simulated Author's Rebuttal

1 responses · 1 unresolved

We thank the referee for the detailed review and constructive comment on the abstract. We agree that the key similarity assumption requires stronger empirical support in the manuscript to substantiate the derivation claims. Below we address the major comment point by point.

read point-by-point responses

  1. Referee: Abstract: The claim that KCLA is 'directly derived from the foundational definition of layer attention' and achieves linear complexity while preserving dynamic updates hinges on the high cosine similarity of keys. The abstract presents this similarity as an empirical observation without providing quantitative analysis, bounds, or proof that it holds independently of network depth or task, which is necessary to confirm the derivation is exact rather than approximate.

    Authors: We agree that the abstract would benefit from quantitative support for the key cosine similarity observation. In the revised version, we will add a new figure and table in Section 3 (or a dedicated subsection) reporting average cosine similarities between keys across consecutive layers, computed on ImageNet, COCO, and medical segmentation datasets for networks of varying depths (e.g., 12 to 50 layers). This will include mean, std, and min/max values to demonstrate robustness. The derivation itself remains exact under the modeling choice of replacing the full key matrix with its highly correlated approximation; the similarity is the enabling empirical condition rather than an approximation in the math. We cannot supply theoretical bounds or a proof of similarity holding for arbitrary depth/task, as this is an observed property of trained transformers rather than a universal guarantee. revision: partial

standing simulated objections not resolved
  • A mathematical proof or theoretical bounds that would guarantee high key cosine similarity independently of network depth or task.

Circularity Check

0 steps flagged

No significant circularity; derivation presented as independent of empirical observation

full rationale

The paper states that KCLA 'achieves linear computational complexity while preserving dynamic information updates, directly derived from the foundational definition of layer attention' after noting an empirical observation of high key cosine similarity as inspiration. No equations or steps are shown that reduce the linear form to the similarity observation by construction, nor are there self-citations, fitted inputs renamed as predictions, or uniqueness theorems invoked. The central claim is framed as following from the layer attention definition itself, rendering the derivation self-contained against external benchmarks with no load-bearing reduction to inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Only the abstract is available, so the ledger is limited to the explicit observation stated in the text.

axioms (1)
  • domain assumption Key representations in layer attention exhibit high cosine similarity.
    This observation is invoked to derive the linear form of KCLA.
invented entities (1)
  • Key-Correlated Layer Attention (KCLA) no independent evidence
    purpose: Linear-complexity replacement for standard layer attention.
    New attention variant introduced in the paper.

pith-pipeline@v0.9.1-grok · 5706 in / 1192 out tokens · 31159 ms · 2026-06-30T01:07:16.412515+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

58 extracted references · 14 canonical work pages · 8 internal anchors

  1. [1]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,”Advances in neural informa- tion processing systems, 2017

    2017

  2. [2]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in2016 IEEE Con- ference on Computer Vision and Pattern Recognition (CVPR), 2016

    2016

  3. [3]

    Densely connected convolutional networks,

    G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Wein- berger, “Densely connected convolutional networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017

    2017

  4. [4]

    On a sparse shortcut topology of artificial neural networks,

    F.-L. Fan, D. Wang, H. Guo, Q. Zhu, P. Yan, G. Wang, and H. Yu, “On a sparse shortcut topology of artificial neural networks,”IEEE Transactions on Artificial Intel- ligence, 2021

    2021

  5. [5]

    Devel- opment of skip connection in deep neural networks for computer vision and medical image analysis: A survey,

    G. Xu, X. Wang, X. Wu, X. Leng, and Y . Xu, “Devel- opment of skip connection in deep neural networks for computer vision and medical image analysis: A survey,” arXiv preprint arXiv:2405.01725, 2024

    work page arXiv 2024

  6. [6]

    Recurrence along depth: Deep convolutional neural networks with recur- rent layer aggregation,

    J. Zhao, Y . Fang, and G. Li, “Recurrence along depth: Deep convolutional neural networks with recur- rent layer aggregation,” inAdvances in Neural Informa- tion Processing Systems, M. Ranzato, A. Beygelzimer, Y . Dauphin, P. Liang, and J. W. Vaughan, Eds. Curran Associates, Inc., 2021

    2021

  7. [7]

    Cross-layer retrospective retrieving via layer attention,

    Y . Fang, Y . Cai, J. Chen, J. Zhao, G. Tian, and G. Li, “Cross-layer retrospective retrieving via layer attention,” 2023

    2023

  8. [8]

    Strengthening layer interaction via dynamic layer attention,

    K. Wang, X. Xia, J. Liu, Z. Yi, and T. He, “Strengthening layer interaction via dynamic layer attention,”arXiv preprint arXiv:2406.13392, 2024

    work page arXiv 2024

  9. [9]

    Linformer: Self-Attention with Linear Complexity

    S. Wang, B. Z. Li, M. Khabsa, H. Fang, and H. Ma, “Linformer: Self-attention with linear complexity,”arXiv preprint arXiv:2006.04768, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2006

  10. [10]

    Hydra attention: Efficient attention with many heads,

    D. Bolya, C.-Y . Fu, X. Dai, P. Zhang, and J. Hoffman, “Hydra attention: Efficient attention with many heads,” inEuropean Conference on Computer Vision. Springer, 2022

    2022

  11. [11]

    Transformer-vq: Linear-time transformers via vector quantization,

    L. D. Lingle, “Transformer-vq: Linear-time transformers via vector quantization,”arXiv preprint arXiv:2309.16354, 2023

    work page arXiv 2023

  12. [12]

    Gated linear attention transformers with hardware- efficient training,

    S. Yang, B. Wang, Y . Shen, R. Panda, and Y . Kim, “Gated linear attention transformers with hardware- efficient training,” 2024

    2024

  13. [13]

    A bidirectional feedforward neural network architecture using the discretized neural memory ordinary differential equation,

    H. Niu, Z. Yi, and T. He, “A bidirectional feedforward neural network architecture using the discretized neural memory ordinary differential equation,”International Journal of Neural Systems, vol. 34, no. 04, p. 2450015, 2024

    2024

  14. [14]

    Cnm-unet: Continuous ordinary differential equations for medical image segmentation,

    T. Xu, Y . Zhu, Q. He, Y . Cao, K. Wang, Z. Yi, and T. He, “Cnm-unet: Continuous ordinary differential equations for medical image segmentation,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 14, 2026, pp. 11 406–11 414

    2026

  15. [15]

    Mobileode: An extra lightweight network,

    L. Yu, J. Wu, B. Gou, X. Min, L. Zhang, Z. Yi, and T. He, “Mobileode: An extra lightweight network,”Advances in Neural Information Processing Systems, vol. 38, pp. 120 931–120 956, 2026

    2026

  16. [16]

    A lightweight u-like network utilizing neural memory ordinary differen- tial equations for slimming the decoder,

    Q. He, X. Yao, J. Wu, Z. Yi, and T. He, “A lightweight u-like network utilizing neural memory ordinary differen- tial equations for slimming the decoder,” inProceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, 2024, pp. 821–829

    2024

  17. [17]

    En- hancing feature fusion of u-like networks with dynamic skip connections,

    Y . Cao, Q. He, K. Wang, J. Xiong, Z. Yi, and T. He, “En- hancing feature fusion of u-like networks with dynamic skip connections,”Medical Image Analysis, p. 104010, 2026. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 11

    2026

  18. [18]

    Subtraction gates: Another way to learn long-term dependencies in recurrent neural networks,

    T. He, H. Mao, and Z. Yi, “Subtraction gates: Another way to learn long-term dependencies in recurrent neural networks,”IEEE Transactions on Neural Networks and Learning Systems, vol. 33, no. 4, pp. 1740–1751, 2020

    2020

  19. [19]

    Cascade-refine model for cephalometric land- mark detection in high-resolution orthodontic images,

    T. He, J. Guo, W. Tang, W. Zeng, P. He, F. Zeng, and Z. Yi, “Cascade-refine model for cephalometric land- mark detection in high-resolution orthodontic images,” Knowledge-Based Systems, vol. 265, p. 110332, 2023

    2023

  20. [20]

    U-net: Convo- lutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convo- lutional networks for biomedical image segmentation,” in International Conference on Medical Image Computing and Computer-Assisted Intervention, 2015

    2015

  21. [21]

    Why is everyone training very deep neural network with skip connections?

    O. Oyedotun, K. Al Ismaeil, and D. Aouada, “Why is everyone training very deep neural network with skip connections?”IEEE Transactions on Neural Networks and Learning Systems, 2021

    2021

  22. [22]

    Dianet: Dense-and-implicit attention network,

    Z. Huang, S. Liang, M. Liang, and H. Yang, “Dianet: Dense-and-implicit attention network,” 2019

    2019

  23. [23]

    Strengthening layer interaction via global-layer attention network for image classification,

    J. You and K. Tang, “Strengthening layer interaction via global-layer attention network for image classification,” in2024 9th International Conference on Intelligent Com- puting and Signal Processing (ICSP), 2024

    2024

  24. [24]

    From layers to states: A state space model perspective to deep neural network layer dynamics,

    Q. Liu, W. Zhao, W. Huang, Y . Fang, L. Yu, and G. Li, “From layers to states: A state space model perspective to deep neural network layer dynamics,”arXiv preprint arXiv:2502.10463, 2025

    work page arXiv 2025

  25. [25]

    Residual Connections Harm Generative Representation Learning

    X. Zhang, R. Jiang, W. Gao, R. Willett, and M. Maire, “Residual connections harm generative representation learning,”arXiv preprint arXiv:2404.10947, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  26. [26]

    Mud- dformer: Breaking residual bottlenecks in transformers via multiway dynamic dense connections,

    D. Xiao, Q. Meng, S. Li, and X. Yuan, “Mud- dformer: Breaking residual bottlenecks in transformers via multiway dynamic dense connections,”arXiv preprint arXiv:2502.12170, 2025

    work page arXiv 2025

  27. [27]

    Enhancing Layer Attention Efficiency through Pruning Redundant Retrievals

    H. Li and X. Huang, “Enhancing layer attention ef- ficiency through pruning redundant retrievals,”arXiv preprint arXiv:2503.06473, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  28. [28]

    Transformers are rnns: Fast autoregressive transformers with linear attention,

    A. Katharopoulos, A. Vyas, N. Pappas, and F. Fleuret, “Transformers are rnns: Fast autoregressive transformers with linear attention,” 2020

    2020

  29. [29]

    Flatten transformer: Vision transformer using focused linear at- tention,

    D. Han, X. Pan, Y . Han, S. Song, and G. Huang, “Flatten transformer: Vision transformer using focused linear at- tention,” inProceedings of the IEEE/CVF international conference on computer vision, 2023

    2023

  30. [30]

    Reformer: The Efficient Transformer

    N. Kitaev, Ł. Kaiser, and A. Levskaya, “Reformer: The efficient transformer,”arXiv preprint arXiv:2001.04451, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2001

  31. [31]

    Attention temperature matters in vit-based cross-domain few-shot learning,

    Y . Zou, R. Ma, Y . Li, and R. Li, “Attention temperature matters in vit-based cross-domain few-shot learning,” inAdvances in Neural Information Processing Systems. Curran Associates, Inc., 2024

    2024

  32. [32]

    Mitigating hallucinations in diffusion models through adaptive at- tention modulation,

    T. Oorloff, Y . Yacoob, and A. Shrivastava, “Mitigating hallucinations in diffusion models through adaptive at- tention modulation,”arXiv preprint arXiv:2502.16872, 2025

    work page arXiv 2025

  33. [33]

    Random search for hyper- parameter optimization,

    J. Bergstra and Y . Bengio, “Random search for hyper- parameter optimization,”The journal of machine learn- ing research, 2012

    2012

  34. [34]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009

    2009

  35. [35]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, “Learning multiple layers of features from tiny images,” 2009

    2009

  36. [36]

    Squeeze- and-excitation networks,

    J. Hu, L. Shen, S. Albanie, G. Sun, and E. Wu, “Squeeze- and-excitation networks,” 2019

    2019

  37. [37]

    Eca- net: Efficient channel attention for deep convolutional neural networks,

    Q. Wang, B. Wu, P. Zhu, P. Li, W. Zuo, and Q. Hu, “Eca- net: Efficient channel attention for deep convolutional neural networks,” 2020

    2020

  38. [38]

    Ba-net: Bridge attention for deep convolutional neural networks,

    Y . Zhao, J. Chen, Z. Zhang, and R. Zhang, “Ba-net: Bridge attention for deep convolutional neural networks,” inComputer Vision – ECCV 2022. Springer Nature Switzerland, 2022

    2022

  39. [39]

    Non- local neural networks,

    X. Wang, R. Girshick, A. Gupta, and K. He, “Non- local neural networks,” in2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

    2018

  40. [40]

    Gcnet: Non- local networks meet squeeze-excitation networks and beyond,

    Y . Cao, J. Xu, S. Lin, F. Wei, and H. Hu, “Gcnet: Non- local networks meet squeeze-excitation networks and beyond,” inProceedings of the IEEE/CVF international conference on computer vision workshops, 2019

    2019

  41. [41]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” 2018

    2018

  42. [42]

    Attention augmented convolutional networks,

    I. Bello, B. Zoph, Q. Le, A. Vaswani, and J. Shlens, “Attention augmented convolutional networks,” in2019 IEEE/CVF International Conference on Computer Vision (ICCV), 2019

    2019

  43. [43]

    Microsoft coco: Common objects in context,

    T.-Y . Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Doll ´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” inComputer vision– ECCV 2014. Springer, 2014

    2014

  44. [44]

    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. Xuet al., “Mmdetection: Open mmlab detection toolbox and benchmark,”arXiv preprint arXiv:1906.07155, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1906

  45. [45]

    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,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017

    2017

  46. [46]

    Mask r-cnn,

    K. He, G. Gkioxari, P. Doll ´ar, and R. Girshick, “Mask r-cnn,” inProceedings of the IEEE international confer- ence on computer vision, 2017

    2017

  47. [47]

    A multi-scale transformer for medical image segmentation: Architectures, model efficiency, and benchmarks,

    Y . Gao, M. Zhou, D. Liu, and D. N. Metaxas, “A multi-scale transformer for medical image segmentation: Architectures, model efficiency, and benchmarks,”ArXiv, 2022

    2022

  48. [48]

    Transfuse: Fusing trans- formers and cnns for medical image segmentation,

    Y . Zhang, H. Liu, and Q. Hu, “Transfuse: Fusing trans- formers and cnns for medical image segmentation,” in Medical image computing and computer assisted inter- vention. Springer, 2021

    2021

  49. [49]

    Unext: Mlp- based rapid medical image segmentation network,

    J. M. J. Valanarasu and V . M. Patel, “Unext: Mlp- based rapid medical image segmentation network,” in International conference on medical image computing and computer-assisted intervention. Springer, 2022

    2022

  50. [50]

    Malunet: A multi-attention and light-weight unet for skin lesion segmentation,

    J. Ruan, S. Xiang, M. Xie, T. Liu, and Y . Fu, “Malunet: A multi-attention and light-weight unet for skin lesion segmentation,” in2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). IEEE, JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 12 2022

    2021

  51. [51]

    Unet++: A nested u-net architecture for med- ical image segmentation,

    Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, “Unet++: A nested u-net architecture for med- ical image segmentation,” inDeep learning in medical image analysis and multimodal learning for clinical decision support. Springer, 2018

    2018

  52. [52]

    Shallow attention network for polyp segmen- tation,

    J. Wei, Y . Hu, R. Zhang, Z. Li, S. K. Zhou, and S. Cui, “Shallow attention network for polyp segmen- tation,” inMedical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part I 24. Springer, 2021

    2021

  53. [53]

    N. C. F. Codella, D. Gutman, M. E. Celebi, B. Helba, M. A. Marchetti, S. W. Dusza, A. Kalloo, K. Liopyris, N. Mishra, H. Kittler, and A. Halpern, “Skin lesion analy- sis toward melanoma detection: A challenge at the 2017 international symposium on biomedical imaging (isbi), hosted by the international skin imaging collaboration (isic),” in2018 IEEE 15th I...

    2017

  54. [54]

    Skin Lesion Analysis Toward Melanoma Detection 2018: A Challenge Hosted by the International Skin Imaging Collaboration (ISIC)

    N. Codella, V . Rotemberg, P. Tschandl, M. E. Celebi, S. Dusza, D. Gutman, B. Helba, A. Kalloo, K. Liopy- ris, M. Marchettiet al., “Skin lesion analysis toward melanoma detection 2018: A challenge hosted by the international skin imaging collaboration (isic),”arXiv preprint arXiv:1902.03368, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  55. [55]

    Kvasir- seg: A segmented polyp dataset,

    D. Jha, P. H. Smedsrud, M. A. Riegler, P. Halvorsen, T. de Lange, D. Johansen, and H. D. Johansen, “Kvasir- seg: A segmented polyp dataset,” inMultiMedia Model- ing: 26th International Conference, 2020

    2020

  56. [56]

    Ege-unet: an efficient group enhanced unet for skin lesion segmen- tation,

    J. Ruan, M. Xie, J. Gao, T. Liu, and Y . Fu, “Ege-unet: an efficient group enhanced unet for skin lesion segmen- tation,” inInternational conference on medical image computing and computer-assisted intervention. Springer, 2023

    2023

  57. [57]

    Decoupled Weight Decay Regularization

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,”arXiv preprint arXiv:1711.05101, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  58. [58]

    SGDR: Stochastic Gradient Descent with Warm Restarts

    ——, “Sgdr: Stochastic gradient descent with warm restarts,”arXiv preprint arXiv:1608.03983, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016