Pith. sign in

REVIEW 4 major objections 6 minor 36 references

The Power of Certainty: How Confident Models Lead to Better Segmentation

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

Pith's one-line read A training-time confidence weight on self-distillation consistency losses improves polyp segmentation and generalization across datasets at zero extra test-time cost.

desk verdict A cheap, plausible confidence-weighted self-distillation trick that shows real but inconsistent gains—worth reviewing, but the paper's loss is underspecified and the SOTA claims overreach. read the letter →

arxiv 2507.10490 v1 pith:XO72GBTN submitted 2025-07-14 cs.CV

classification cs.CV
keywords Self-distillationConfidence-baseddistillationRegularizationPolypsegmentationMedicalimageConvolutionalnetworksDynamicconfidenceGeneralization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes Dynamic Confidence-Based Self-Distillation (DCSD), a training-time regularizer that adds a consistency term between the model's predictions in consecutive batches, weighted by how well the earlier prediction matched the ground truth. The central claim is that this simple weighting improves segmentation accuracy and, more importantly, generalization to unseen datasets, while adding no extra computation or memory at test time. Experiments on polyp segmentation across multiple colonoscopy and wireless endoscopy datasets support that claim: the weighted version beats both a plain model and conventional self-distillation on the held-out data_c6 set and on unseen EndoScene, ETIS, and BKAI-IGH sets.

What carries the argument

The dynamic confidence coefficient is the mechanism. It is the Dice score between the previous iteration's softened prediction and the ground truth, used as a multiplicative weight on the MSE consistency term in the DCSD loss. High-Dice predictions are treated as trustworthy targets and enforced, while low-Dice predictions are down-weighted. Temperature softening (T=4) is applied to obtain soft confidence scores, and only the previous mini-batch's predictions are retained.

What would settle it

Train the same architecture on a dataset with artificially corrupted labels (e.g., 30% flipped masks). If DCSD's confidence weighting works as claimed, the corrupted high-Dice predictions should be down-weighted and the model should degrade gracefully; if instead DCSD amplifies those errors and yields lower Dice than base training, then the confidence mechanism is not performing the claimed reweighting.

Watch

Extended reading notes

Core claim

The central discovery is that a dynamic confidence coefficient, computed as the Dice overlap between the previous iteration's temperature-softened prediction and the ground truth, can reweight the mean-squared-error consistency loss between consecutive batch predictions, and doing so consistently improves performance over both base training and unweighted self-distillation. The DCSD loss is $\mathcal{L}_{DCSD} = \frac{1}{n}\sum_{i=1}^{n} \text{Dice}(p^{t-1}_i, y^{t-1}_i) \cdot \text{MSE}(p^t_i, p^{t-1}_i)$. Only the previous mini-batch's predictions need to be stored during training; no teacher model or extra inference cost is required.

Load-bearing premise

The load-bearing premise is that a prediction's Dice against the ground truth in the previous iteration is a reliable signal of which predictions are safe to use as distillation targets; if early predictions are miscalibrated or the batch is small, the weighting can reinforce errors instead of preventing them.

Editorial extensions

If this is right

  • If DCSD is correct, any segmentation model can gain accuracy and cross-dataset generalization by adding a few lines of training code: reweight the consistency loss by the previous prediction's Dice with the label.
  • The improvement transfers across architectures: both established polyp segmentation models improved with DCSD over their base and self-distilled versions, so the regularizer is not tied to the proposed encoder-decoder backbone.
  • The method appears to generalize to unseen datasets: DCSD raised Dice to 89.54 on EndoScene, 71.21 on ETIS, and 81.41 on BKAI-IGH compared with base and self-distillation baselines.
  • Because no extra computation or memory is used during inference, the approach could be deployed in real-time colonoscopy settings without changing runtime.

Reading between the lines

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

  • A natural extension is to apply DCSD to other dense prediction tasks and to noisy-label regimes; the confidence weighting may act as a self-paced regularizer that naturally down-weights unreliable early predictions.
  • The dependence on prior-iteration Dice with ground truth suggests that batch size and label quality are critical: with very small batches the confidence estimate is noisy, and with mislabeled ground truth the weight can systematically amplify wrong targets, an effect the paper does not quantify.
  • The theoretical argument that confidence-based regularization shrinks the hypothesis space is asserted rather than derived; a falsifiable test would be to measure empirical complexity, such as sharpness or effective capacity, with and without DCSD.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 proposes Dynamic Confidence-based Self-Distillation (DCSD), a training-time regularizer for polyp segmentation. DCSD computes a consistency loss between the current iteration's predictions and the previous mini-batch's softened predictions, weighted by a 'dynamic confidence coefficient' derived from the Dice overlap of the previous prediction with the ground truth. The authors combine DCSD with a PVT-backed encoder-decoder architecture with RFB and layer aggregation. They report that DCSD outperforms the base model and conventional self-distillation on the data_c6 test set, and on three of six cross-dataset targets (EndoScene, ETIS, BKAI-IGH), with no extra test-time cost. The paper also includes ablations on soft versus hard confidence and a comparison against ten published polyp segmentation methods.

Significance. If DCSD is exactly as described and the gains are reproducible, it would be a low-cost regularizer applicable to any segmentation model, since it only stores the previous mini-batch and has no inference overhead. The paper's three-architecture ablation (TransNetR, ShallowNet, and the proposed model) is a useful check on generality, and the cross-dataset experiments address an important clinical need. However, the manuscript currently defines the loss in two contradictory ways, omits a hyperparameter, and overstates the state-of-the-art comparison, so the empirical contribution cannot be assessed until these issues are resolved.

major comments (4)
  1. [Section 3.1, Eq. (1); Algorithm 1 line 13; Algorithm 2 lines 5-6] The loss definition is contradictory. Eq. (1) defines L_DCSD = (1/n) Σ Dice(p^{t-1}_i, y^{t-1}_i) · MSE(p^t_i, p^{t-1}_i), so a high-Dice previous prediction receives a larger weight. Algorithm 2, lines 5–6, sets confidence-coefficient = 1 − dice_loss(pre_out, pre_label) and loss = consistency · confidence-coefficient, so a high-Dice previous prediction receives a smaller weight. These are opposite weighting schemes, and both are called 'dynamic confidence.' Algorithm 1, line 12, calls dcsd(out_pre, pre_out, pre_label) with an argument order that does not match Eq. (1), and line 13 multiplies by an undefined hyperparameter t2 whose value is never given in Section 4.1. Because the sign and form of the confidence weight are the entire mechanism of the method, the reported improvements in Tables 3–5 cannot be attributed to a well-defined DCSD loss as it stands.
  2. [Section 3.2] The theoretical analysis asserts Complexity(H_DCSD) < Complexity(H_SD) without defining either hypothesis space or proving the inequality. Since DCSD changes only the training objective and not the model family or parameterization, the hypothesis space of the trained model is not obviously smaller than that of the base model, so the claimed tighter generalization bound does not follow. This section should be removed or replaced with a rigorous argument, or explicitly labeled as an intuition rather than a theorem.
  3. [Section 5.1 and Abstract] The abstract states that DCSD 'outperforms state-of-the-art models,' but Table 2 shows that Ours ties DeepLabV3+ (ResNet101) on both Dice (0.82) and IoU (0.75) and has lower precision (0.91 vs. 0.92). The only metric on which Ours is strictly better is Recall (0.82 vs. 0.81). The claim should be toned down to 'matches or exceeds' or the comparison should be expanded to show a statistically meaningful improvement.
  4. [Section 5.2, Table 4] The claim that DCSD 'achieves superior results' on cross-dataset generalization is not consistent across Table 4: DCSD is worse than SD on Kvasir (0.8985 vs. 0.9022 Dice), ClinicDB (0.8994 vs. 0.9036), and ColonDB (0.7639 vs. 0.7681). The paper reports a single run without confidence intervals or significance tests, so the gains on EndoScene, ETIS, and BKAI-IGH cannot be distinguished from noise. Please report multiple seeds and statistical testing, and discuss the datasets where DCSD underperforms SD.
minor comments (6)
  1. [Section 5.3] The word 'temparature' should be 'temperature'.
  2. [Equation (1)] Equation (1) uses the same index i for p^t_i and p^{t-1}_i, but Algorithm 1 compares predictions on the previous batch's images, so the indices and variables should be defined to match the algorithm.
  3. [Algorithm 2] The input description in Algorithm 2 says 'the previous mini-batch's prediction ... is denoted as pre_out, while the current iteration's prediction ... is denoted as out_pre'; this naming conflicts with Algorithm 1 and should be aligned with Eq. (1).
  4. [Figure 1 caption] The caption does not define the symbols B_t, θ_t, and p_t in the text; please spell out the notation.
  5. [Section 4.1] The distillation weight t2 and the temperature T used in the main experiments are not specified in the experimental setup; T=4 appears only in Table 5, and t2 is never given a value.
  6. [Section 1] The introduction misspells 'Radcliffe' as 'Radclife'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: DCSD is a training-time consistency loss whose confidence weight is computed from the current batch's ground truth and previous-iteration predictions, not from the reported test metrics or from a fitted parameter; the central performance claims do not reduce to the method's inputs.

full rationale

DCSD (Eq. 1) is L_DCSD = (1/n) sum_i Dice(p^{t-1}_i, y^{t-1}_i) * MSE(p^t_i, p^{t-1}_i). The confidence coefficient is computed online from the previous training batch's ground truth and the model's own softened previous prediction; it is not a fitted parameter and is not evaluated on the test set. The reported Dice/IoU gains on data_c6, EndoScene, ETIS and BKAI-IGH are empirical training-time effects, and no equation in the paper defines those reported test metrics as an input to the loss or as the thing being optimized. There are no load-bearing self-citations: the cited priors are external, and the claimed benefit of confidence weighting is not imported from any cited uniqueness or optimality theorem. Section 3.2's complexity-reduction argument is asserted rather than derived, but an unsupported generalization-bound heuristic is a correctness concern, not a circular reduction. The genuine manuscript problems are reproducibility and evaluation, not circularity: Algorithm 2 sets confidence-coefficient = 1 - Dice(pre_out, pre_label), which is the opposite weighting to Eq. 1; Algorithm 1 uses an undefined t2 and swaps argument order; and Section 5.3 selects temperature T=4 using the data_c6 test set before the same data_c6 comparison appears in Tables 2-3. These flaws make the exact method ambiguous and can inflate the headline numbers, but they do not make any prediction equivalent to its input by construction.

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

The paper introduces one new loss formulation and combines existing modules (PVT, RFB, layer aggregation). No new physical entities are postulated. The free parameters are the temperature T and the undefined distillation weight t2; the axioms are the standard generalization bound, the unproven hypothesis-space reduction claim, and the confidence-as-reliability assumption.

free parameters (2)
  • Temperature T for softening predictions = T=4 (selected via ablation on data_c6, Table 5)
    The temperature parameter softens the previous prediction before computing the confidence score. The paper's default value T=4 is justified by an ablation on the data_c6 test set.
  • Distillation loss weight t2 in Algorithm 1 = not specified
    Algorithm 1 line 13 uses an undefined 't2' multiplier on the DCSD loss. The value is not stated anywhere in the paper, so the exact training objective cannot be reconstructed.
assumptions (3)
  • standard math Standard PAC-style generalization bound with hypothesis space complexity term (Section 3.2)
    The paper invokes a generic bound R(h) <= R_hat(h) + O(Complexity(H)/sqrt(n)) to motivate the method. The bound is not stated with full technical precision and is used only qualitatively.
  • ad hoc to paper Confidence-weighted distillation reduces the hypothesis space complexity (Complexity(H_DCSD) < Complexity(H_SD))
    Section 3.2 asserts this inequality without derivation. It is the load-bearing premise for the theoretical claim that DCSD has a tighter generalization bound, but no specific hypothesis-space measure is identified or proven to shrink.
  • domain assumption Dice between the previous iteration's prediction and ground truth is a reliable confidence measure for weighting consistency
    The method in Eq. 1 and Algorithm 2 assumes that a higher Dice with the previous batch's ground truth means the previous soft predictions are trustworthy distillation targets. This is an empirical modeling assumption and is not validated independently.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Power of Certainty: How Confident Models Lead to Better Segmentation." pith.science (2026). https://pith.science/paper/XO72GBTN

@misc{pith2026250710490,
  author       = {Pith},
  title        = {Pith review of: The Power of Certainty: How Confident Models Lead to Better Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XO72GBTN}},
  note         = {Machine review of arXiv:2507.10490}
}
read the original abstract

Deep learning models have been proposed for automatic polyp detection and precise segmentation of polyps during colonoscopy procedures. Although these state-of-the-art models achieve high performance, they often require a large number of parameters. Their complexity can make them prone to overfitting, particularly when trained on biased datasets, and can result in poor generalization across diverse datasets. Knowledge distillation and self-distillation are proposed as promising strategies to mitigate the limitations of large, over-parameterized models. These approaches, however, are resource-intensive, often requiring multiple models and significant memory during training. We propose a confidence-based self-distillation approach that outperforms state-of-the-art models by utilizing only previous iteration data storage during training, without requiring extra computation or memory usage during testing. Our approach calculates the loss between the previous and current iterations within a batch using a dynamic confidence coefficient. To evaluate the effectiveness of our approach, we conduct comprehensive experiments on the task of polyp segmentation. Our approach outperforms state-of-the-art models and generalizes well across datasets collected from multiple clinical centers. The code will be released to the public once the paper is accepted.

Figures

Figures reproduced from arXiv: 2507.10490 by the authors.

Figure 1
Figure 1. The figure demonstrates our proposed DCSD (Dynamic Confidence-Based Self-Distillation) approach in detail. DCSD calculates the loss between the previous and the current iterations within a batch using a dynamic confidence coefficient. 𝐵𝑡 , 𝜃𝑡 , and 𝑝 𝑡 represent the batch, model weights, and prediction at the 𝑡-th iteration, respectively [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The overview of our architecture and the novel DCSD approach. PVTV2 represents "Pyramid Vision Transformer" backbone which reduces computational cost through progressively smaller feature map sizes while enhancing generalization across different image sizes and resolutions via multi-scale feature extraction. RFB modules represent Receptive Field Block. Dense aggregation represents deep layer aggregation which aggreg… view at source ↗
Figure 3
Figure 3. Comparison of model outputs from TransNetR, ShallowNet, our model, our model with SD and our model with the proposed DCSD method. The figure highlights the differences in segmentation performance on the data_c6 dataset. estimation to enhance the robustness and transferability of DCSD in varying clinical contexts. References [1] World Health Organization (WHO), Colorectal Cancer. https://www. who.int/news-room/fact-s… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 28 canonical work pages

  1. [1]

    who.int/news-room/fact-sheets/detail/colorectal-cancer, 2023

    World Health Organization (WHO), Colorectal Cancer.https://www. who.int/news-room/fact-sheets/detail/colorectal-cancer, 2023

  2. [2]

    Kvasir- seg: A segmented polyp dataset

    Debesh Jha, Pia H Smedsrud, Michael A Riegler, Pål Halvorsen, Thomas De Lange, Dag Johansen, and Håvard D Johansen. Kvasir- seg: A segmented polyp dataset. InMultiMedia modeling: 26th in- ternational conference, MMM 2020, Daejeon, South Korea, January 5–8, 2020, proceedings, part II 26, pages 451–462. Springer, 2020

  3. [3]

    Pyramid vision transformer: A versatile backbone for dense prediction without con- volutions

    Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without con- volutions. InProceedings of the IEEE/CVF international conference on computer vision, pages 568–578, 2021

  4. [4]

    A multi-centre polyp detection and segmentation dataset for generalisability assessment

    Sharib Ali, Debesh Jha, Noha Ghatwary, Stefano Realdon, Renato Cannizzaro, Osama E Salem, Dominique Lamarque, Christian Daul, Michael A Riegler, Kim V Anonsen, et al. A multi-centre polyp detection and segmentation dataset for generalisability assessment. Scientific Data, 10(1):75, 2023

  5. [5]

    Transnetr: Transformer- basedresidualnetworkforpolypsegmentationwithmulti-centerout- of-distribution testing

    D.Jha, N.Tomar, V.Sharma, and U.Bagci. Transnetr: Transformer- basedresidualnetworkforpolypsegmentationwithmulti-centerout- of-distribution testing. InProceedings of the Medical Imaging with Deep Learning, 2023

  6. [6]

    Shallow attention network for polyp segmentation

    Jun Wei, Yiwen Hu, Ruimao Zhang, Zhen Li, S Kevin Zhou, and Shuguang Cui. Shallow attention network for polyp segmentation. In Medical Image Computing and Computer Assisted Intervention– MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part I 24, pages 699–

  7. [7]

    Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs

    Jorge Bernal, F Javier Sánchez, Gloria Fernández-Esparrach, Debora Gil, Cristina Rodríguez, and Fernando Vilariño. Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency mapsfromphysicians. Computerizedmedicalimagingandgraphics , 43:99–111, 2015

  8. [8]

    A benchmark for endoluminal scene segmen- tation of colonoscopy images

    David Vázquez, Jorge Bernal, F Javier Sánchez, Gloria Fernández- Esparrach, Antonio M López, Adriana Romero, Michal Drozdzal, and Aaron Courville. A benchmark for endoluminal scene segmen- tation of colonoscopy images. Journal of healthcare engineering, 2017(1):4037190, 2017. First Author et al.:Preprint submitted to Elsevier Page 8 of 9 Short Title of th...

Show all 36 references
  1. [9]

    Toward embedded detection of polyps in wce imagesforearlydiagnosisofcolorectalcancer

    Juan Silva, Aymeric Histace, Olivier Romain, Xavier Dray, and Bertrand Granado. Toward embedded detection of polyps in wce imagesforearlydiagnosisofcolorectalcancer. Internationaljournal of computer assisted radiology and surgery, 9:283–293, 2014

  2. [10]

    Neounet: Towards accurate colon polyp segmentation and neoplasm detection

    Phan Ngoc Lan, Nguyen Sy An, Dao Viet Hang, Dao Van Long, TranQuangTrung,NguyenThiThuy,andDinhVietSang. Neounet: Towards accurate colon polyp segmentation and neoplasm detection. In Advances in visual computing: 16th international symposium, ISVC 2021, virtual event, October 4...

  3. [11]

    Auto- matedpolypdetectionincolonoscopyvideosusingshapeandcontext information

    Nima Tajbakhsh, Suryakanth R Gurudu, and Jianming Liang. Auto- matedpolypdetectionincolonoscopyvideosusingshapeandcontext information. IEEE transactions on medical imaging, 35(2):630–644, 2015

  4. [12]

    U-net: Con- volutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Con- volutional networks for biomedical image segmentation. InMedical imagecomputingandcomputer-assistedintervention–MICCAI2015: 18thinternationalconference,Munich,Germany,October5-9,2015, proceedings, part III 18, pag...

  5. [13]

    Attentionu-net:Learningwheretolookforthepancreas

    OzanOktay. Attentionu-net:Learningwheretolookforthepancreas. arXiv preprint arXiv:1804.03999, 2018

  6. [14]

    Unet++:Anestedu-netarchitectureformedical image segmentation

    Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, andJianmingLiang. Unet++:Anestedu-netarchitectureformedical image segmentation. In Deep Learning in Medical Image Analysis andMultimodalLearningforClinicalDecisionSupport:4thInterna- tionalWorkshop,DLMIA2018,and8thInt...

  7. [15]

    Unet3+:Afull-scaleconnectedunetformedicalimagesegmentation

    Huimin Huang, Lanfen Lin, Ruofeng Tong, Hongjie Hu, Qiaowei Zhang, Yutaro Iwamoto, Xianhua Han, Yen-Wei Chen, and Jian Wu. Unet3+:Afull-scaleconnectedunetformedicalimagesegmentation. In ICASSP 2020-2020 IEEE international conference on acoustics, speech and signal processing (...

  8. [16]

    Segnet:A deepconvolutionalencoder-decoderarchitectureforimagesegmenta- tion

    VijayBadrinarayanan,AlexKendall,andRobertoCipolla. Segnet:A deepconvolutionalencoder-decoderarchitectureforimagesegmenta- tion. IEEEtransactionsonpatternanalysisandmachineintelligence , 39(12):2481–2495, 2017

  9. [17]

    Fully convo- lutional networks for semantic segmentation

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convo- lutional networks for semantic segmentation. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 3431–3440, 2015

  10. [18]

    Pyramidsceneparsingnetwork

    Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, and JiayaJia. Pyramidsceneparsingnetwork. In ProceedingsoftheIEEE conference on computer vision and pattern recognition, pages 2881– 2890, 2017

  11. [19]

    Encoder-decoder with atrous separable con- volution for semantic image segmentation

    Liang-ChiehChen,YukunZhu,GeorgePapandreou,FlorianSchroff, and Hartwig Adam. Encoder-decoder with atrous separable con- volution for semantic image segmentation. In Proceedings of the European conference on computer vision (ECCV), pages 801–818, 2018

  12. [20]

    Road extraction bydeepresidualu-net

    Zhengxin Zhang, Qingjie Liu, and Yunhong Wang. Road extraction bydeepresidualu-net. IEEEGeoscienceandRemoteSensingLetters , 15(5):749–753, 2018

  13. [21]

    Cascaded partial decoder for fast and accurate salient object detection

    Zhe Wu, Li Su, and Qingming Huang. Cascaded partial decoder for fast and accurate salient object detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 3907–3916, 2019

  14. [22]

    Pranet: Parallel reverse attention network for polyp segmentation

    Deng-Ping Fan, Ge-Peng Ji, Tao Zhou, Geng Chen, Huazhu Fu, Jianbing Shen, and Ling Shao. Pranet: Parallel reverse attention network for polyp segmentation. In International conference on medical image computing and computer-assisted intervention, pages 263–273. Springer, 2020

  15. [23]

    Automatic polyp segmentation via multi-scale subtraction network

    Xiaoqi Zhao, Lihe Zhang, and Huchuan Lu. Automatic polyp segmentation via multi-scale subtraction network. InMedical Im- age Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27– October 1, 2021, Proceedings,...

  16. [24]

    Hardnet- mseg: A simple encoder-decoder polyp segmentation neural net- work that achieves over 0.9 mean dice and 86 fps.arXiv preprint arXiv:2101.07172, 2021

    Chien-Hsiang Huang, Hung-Yu Wu, and Youn-Long Lin. Hardnet- mseg: A simple encoder-decoder polyp segmentation neural net- work that achieves over 0.9 mean dice and 86 fps.arXiv preprint arXiv:2101.07172, 2021

  17. [25]

    Hardnet: A low memory traffic network

    Ping Chao, Chao-Yang Kao, Yu-Shan Ruan, Chien-Hsiang Huang, and Youn-Long Lin. Hardnet: A low memory traffic network. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3552–3561, 2019

  18. [26]

    Distillingtheknowledgeinaneuralnetwork

    GeoffreyHinton. Distillingtheknowledgeinaneuralnetwork. arXiv preprint arXiv:1503.02531, 2015

  19. [27]

    Fitnets: Hints for thin deep nets.arXiv preprint arXiv:1412.6550, 2014

    Adriana Romero, Nicolas Ballas, Samira Ebrahimi Kahou, Antoine Chassang, Carlo Gatta, and Yoshua Bengio. Fitnets: Hints for thin deep nets.arXiv preprint arXiv:1412.6550, 2014

  20. [28]

    SergeyZagoruykoandNikosKomodakis.Payingmoreattentiontoat- tention:Improvingtheperformanceofconvolutionalneuralnetworks via attention transfer.arXiv preprint arXiv:1612.03928, 2016

  21. [29]

    Relational knowledge distillation

    Wonpyo Park, Dongju Kim, Yan Lu, and Minsu Cho. Relational knowledge distillation. InProceedings of the IEEE/CVF conference oncomputervisionandpatternrecognition ,pages3967–3976,2019

  22. [30]

    Beyourownteacher:Improvetheperformanceof convolutional neural networks via self distillation

    LinfengZhang,JieboSong,AnniGao,JingweiChen,ChenglongBao, andKaishengMa. Beyourownteacher:Improvetheperformanceof convolutional neural networks via self distillation. InProceedings of the IEEE/CVF international conference on computer vision, pages 3713–3722, 2019

  23. [31]

    Born again neural networks

    Tommaso Furlanello, Zachary Lipton, Michael Tschannen, Laurent Itti, and Anima Anandkumar. Born again neural networks. In International conference on machine learning, pages 1607–1616. PMLR, 2018

  24. [32]

    Self-distillation from the last mini-batch for consistency regulariza- tion

    Yiqing Shen, Liwu Xu, Yuzhe Yang, Yaqian Li, and Yandong Guo. Self-distillation from the last mini-batch for consistency regulariza- tion. InProceedingsoftheIEEE/CVFconferenceoncomputervision and pattern recognition, pages 11943–11952, 2022

  25. [33]

    Receptive field block net for accurate andfastobjectdetection

    Songtao Liu, Di Huang, et al. Receptive field block net for accurate andfastobjectdetection. In ProceedingsoftheEuropeanconference on computer vision (ECCV), pages 385–400, 2018

  26. [34]

    Deep layer aggregation

    FisherYu,DequanWang,EvanShelhamer,andTrevorDarrell. Deep layer aggregation. InProceedings of the IEEE conference on com- puter vision and pattern recognition, pages 2403–2412, 2018

  27. [35]

    Decoupled weight decay regular- ization, 2019

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regular- ization, 2019

  28. [36]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770– 778, 2016. First Author et al.:Preprint submitted to Elsevier Page 9 of 9

Pith tools

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