Pith. sign in

REVIEW 4 major objections 6 minor 46 references

Domain Adaptive Diabetic Retinopathy Grading with Model Absence and Flowing Data

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A VAE trained on saliency maps can adapt a frozen, unseen diabetic-retinopathy model to new hospital data.

desk verdict OMG-DA is a useful new problem statement and the saliency-supervised VAE shows some empirical promise, but the theoretical framing is vacuous and the comparisons don't isolate the model-absence setting. read the letter →

arxiv 2412.01203 v1 pith:YZKKNBB2 submitted 2024-12-02 cs.CV

Wenxin Su , Song Tang , Xiaofeng Liu , Xiaojing Yi , Mao Ye , Chunxiao Zu , Jiahao Li , Xiatian Zhu This is my paper · ORCID
classification cs.CV
keywords diabeticretinopathygradingonlinemodel-agnosticdomainadaptationunadversarialexamplestest-timesource-freevariationalautoencodersaliencymapfundusimageshift
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 proposes a way to adapt a diabetic-retinopathy grading model to a new hospital's images when the model itself is hidden and the new images arrive as a stream—a setting it calls Online Model-aGnostic Domain Adaptation (OMG-DA). Instead of retraining or even touching the model, it rewrites each incoming image into a generative unadversarial example: a small, individualized perturbation that makes the image easier for the hidden model to grade. The perturbation generator is a Variational AutoEncoder trained to regress target images onto their own saliency maps, with the saliency map doing double duty as pseudo-label and theoretical upper bound. Across 12 source-target tasks built from four fundus datasets, the approach raises average accuracy by 4.5% and quadratic weighted kappa by 2.8% over the untouched source model, and it stabilizes existing test-time adaptation methods at small batch sizes. The paper's point is that domain adaptation can be done from the data side alone, with no model access, no source data, and no labels.

What carries the argument

The load-bearing object is the generative unadversarial perturbation function $F_\Phi(\partial \delta_0 / \partial x)$, instantiated as a Variational AutoEncoder. The encoder, with the reparameterization trick, samples the latent variable $z$ that stands for the derivative of the initial random noise with respect to the image, while the decoder generates the individual perturbation $\delta_t$. A by-pass connection adds $\delta_t$ to the input image to form the unadversarial example $\hat{x}_t$, and the training loss is $L_{\mathrm{GUES}} = \alpha D_{\mathrm{KL}}(q(z|x_t) \| \mathcal{N}(0,I)) + \beta \|\hat{x}_t - g_t\|^2$, where $g_t$ is the fine-grained saliency map of $x_t$. The saliency map thus serves simultaneously as the regression target, the carrier of lesion-related signal, and the theoretical upper bound that the paper uses to justify identifying the latent input.

What would settle it

Compute the Jacobian $\partial \delta_0 / \partial x$ for the actual random noise $\delta_0$ used in GUES and check numerically whether its magnitude is bounded by $U \cdot s$ across all pixels and all 12 transfer tasks; a single violation, or a case where the derivative is undefined, would falsify Theorem 2 and remove the theoretical reason for using saliency maps.

Watch

Extended reading notes

Core claim

The central claim is that GUES—a VAE that turns an input fundus image into an individual perturbation by regressing toward the image's fine-grained saliency map—can adapt a completely unseen frozen source model to a streaming target domain. The method reformulates the iterative perturbation optimization of unadversarial learning into a generative form, $\delta_k = \delta_0 + V F_\Phi(\partial \delta_0 / \partial x)$, in which the encoder with reparameterization supplies the latent input and the decoder acts as the perturbation generation function. The saliency map is chosen as the pseudo-perturbation label because it highlights hemorrhages, exudates, and other DR-relevant lesions, and because the paper proves (Theorem 2) that the saliency map bounds the latent input. In experiments on APTOS, DDR, DeepDR, and Messidor-2, GUES outperforms the unaugmented source model by 4.5% average ACC and 2.8% average QWK over 12 transfer tasks, and when prepended to SHOT-IM or TENT it improves those trainable test-time methods as well.

Load-bearing premise

The whole saliency-map choice rests on Theorem 2's claim that the derivative of the initial random noise with respect to the image exists and is bounded by the saliency map; since that noise is drawn independently of the image, the derivative may be zero, undefined, or only boundable under a specific parameterization the paper does not give, and if it fails, the saliency pseudo-labels lose their theoretical justification.

Editorial extensions

If this is right

  • A hospital can deploy a proprietary grading model without revealing its parameters or architecture and still improve its accuracy on the hospital's own incoming images.
  • GUES can be composed with existing trainable test-time adaptation methods such as SHOT-IM and TENT, improving their accuracy and reducing their sensitivity to small batch sizes.
  • Because the generator is trained only on unlabeled target images and their saliency maps, it does not require labels, source data, or model access, which addresses data-privacy and model-privacy constraints in clinical deployment.
  • The visualization analyses suggest the generated perturbations direct the hidden model's attention to hemorrhages, soft exudates, and hard exudates, the features clinicians use for grading.

Reading between the lines

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

  • Editorial extension: if the saliency-bound argument is the only reason for choosing saliency maps, then an ablation with a generic edge map or a random fixed target would separate the bound's role from the VAE's denoising effect; the paper's ablations replace saliency with the input image or a mixup, but not with a non-lesion structural map.
  • Editorial extension: the method's stated limitation implies a natural boundary—on natural images with rich backgrounds the saliency bound constant $U$ becomes too loose, so the same recipe would likely fail outside medical fundus photography; a transfer to chest X-rays or dermoscopy, where backgrounds are more contained, would test how far the data-centric trick generalizes.
  • Editorial extension: the fact that perturbations are individualized rather than class-specific suggests GUES is learning a style or domain translation rather than a label-conditioned shift; comparing GUES perturbations with oracle-trained Grad-CAM maps could reveal whether the hidden model's actual failure modes are being corrected or merely generic contrast enhancement.
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 proposes a new setting, OMG-DA (Online Model-aGnostic Domain Adaptation), in which a frozen, unseen source model is adapted to streaming unlabeled target data without access to the model's parameters or source data. To operate in this setting, the authors introduce GUES, a VAE-based method that generates per-image perturbations by regressing the augmented image onto its fine-grained saliency map. The paper claims two theorems: Theorem 1 reformulates iterative unadversarial perturbation optimization as a generative function of a latent variable (the derivative of initial noise with respect to the input), and Theorem 2 asserts that the saliency map provides an upper bound on this latent variable. Experiments on four diabetic retinopathy datasets across 12 transfer tasks report average improvements of 4.5% ACC and 2.8% QWK over the unaugmented source model, as well as gains when GUES is combined with TTA methods such as SHOT-IM and TENT.

Significance. The OMG-DA setting is clinically motivated and genuinely interesting: model absence, source-data privacy, and streaming target data are all realistic constraints in hospital deployment. The 12-task evaluation across four DR datasets and the batch-size robustness study are useful empirical contributions, and the paper is commendably transparent about its limitations for natural images. However, the claimed theoretical grounding is the central novelty, and it is not sound. If the theory were removed, the method would reduce to an empirical image-to-saliency regression with no demonstrated advantage over simpler model-agnostic input transforms. The empirical evaluation also lacks a model-agnostic control baseline, so the central claims are currently unsupported.

major comments (4)
  1. [Supplementary Sec. 8.1, Eqs. (19)-(20)] Theorem 1 is vacuous as proven. The proof bounds δ_{k+1} by δ0 + U_m[k + ∂δ0/∂x + h_1(∂δ0/∂x) + ... + h_k(∂δ0/∂x)] and then defines F_Φ(∂δ0/∂x) to be exactly the bracketed expression and chooses V 'to make the equality relationship hold.' Because F_Φ and V are constructed after the fact to absorb the upper bound, any iterative sequence can be represented in the form δ_k = δ0 + V·F_Φ(∂δ0/∂x); no substantive generative reformulation is established. This is load-bearing because Section 4.2 uses Theorem 1 to justify replacing iterative optimization with a VAE that predicts perturbations.
  2. [Supplementary Sec. 8.2, Eqs. (23)-(30)] Theorem 2 is not established. In Eq. (10), δ0 is an independently sampled random initialization, not a function of the input x, so ∂δ0/∂x is either zero or undefined. The proof invokes the chain rule ∂δ0/∂x = (∂δ0/∂s)(∂s/∂x), which presupposes that δ0 is a differentiable function of the saliency map s; no such dependency is given. Moreover, Eq. (27) yields ∂s/∂x = 1/2 − Σ_i I_{Δi}/I_Δ in the treated case, a quantity that depends on the perturbation Δx and is not shown to be proportional to the saliency map s. The conclusion ∂s/∂x ∝ s and hence ∂δ0/∂x ≤ U·s is asserted rather than derived. This invalidates the theoretical basis for choosing saliency maps as pseudo-perturbation labels and for the bound used in Section 4.2.
  3. [Sec. 4.2, Eqs. (4)-(8)] The identification of the VAE latent variable z with ∂δ0/∂x is unsupported once Theorems 1 and 2 fail. The claim that the KL term L_KL 'ensures the learning of ∂δ0/∂x' and that the saliency map 'provides an upper bound on the function input' rests entirely on those theorems. As it stands, the method is an empirical regression of x_t + δ_t onto the saliency map g_t; the 'generative unadversarial examples' interpretation and the theoretical motivation for the specific loss are not justified.
  4. [Sec. 5.3 and Table 2] The experimental comparison does not control for the model-absence assumption. The OMG-DA setting forbids access to the source model, yet the main baselines in Table 2 (SHOT, NRC, CoWA, PLUE, TPDS, SHOT-IM, TENT, SAR) access model parameters or outputs. No model-agnostic input-only baseline (e.g., histogram equalization, contrast enhancement, or other image transforms) is included. The reported 4.5% ACC / 2.8% QWK improvement over the unaugmented source model therefore cannot be attributed to GUES's specific mechanism, since any input transformation that shifts the target distribution could produce similar gains.
minor comments (6)
  1. [Sec. 5.2] There is a typo in the subsection title: 'Souce model pre-training' should read 'Source model pre-training.'
  2. [Conclusion] The word 'demostrate' should be 'demonstrate'.
  3. [Sec. 5.5] The subsection title 'Visualization of unadversaisal examples' contains a typo; it should be 'unadversarial examples.'
  4. [Algorithm 1] The algorithm input lists the VAE parameters but not the hyperparameters α and β, and step 6 refers to g_i before defining how the saliency map is computed; adding these details would improve reproducibility.
  5. [Sec. 5.4 and Fig. 4] The batch-size comparison reports point estimates without error bars or significance tests, so the claimed robustness of GUES at small batch sizes is not statistically supported.
  6. [Eq. (31)] The accuracy formula ACC = (TP+TN)/(TP+TN+FP+FN) is written for binary classification, but the DR grading task is multiclass; the macro- or micro-averaging scheme should be specified.

Circularity Check

3 steps flagged · score 8.0 of 10

The theoretical core is circular: Theorem 1 defines FΦ after the fact to force the equality, Theorem 2 assumes a derivative of random noise and asserts ∂s/∂x ∝ s, and the VAE latent is declared to be that derivative.

  1. self definitional [Supplementary Sec. 8.1, Eqs. (19)-(20); used in Sec. 4.1]
    "Let FΦ ( ∂δ0 ∂x ) = ( k + ∂δ0 ∂x + h1 ( ∂δ0 ∂x ) + · · · + hk ( ∂δ0 ∂x ) ) and V be a value that makes the equality relationship hold. Eq. (19) becomes the generative form below. δk = δ0 + V · FΦ ( ∂δ0 ∂x ) . (20)"

    The proof has only established an upper bound δ_{k+1} ≤ δ0 + Um[k + ∂δ0/∂x + Σ h_i(∂δ0/∂x)]. The generative function FΦ is then defined as exactly the bracketed residual, and V is chosen after the fact to convert the inequality into the equality δk = δ0 + V·FΦ(∂δ0/∂x). This makes Theorem 1 true by construction for any iterative sequence, so it cannot ground the paper's inference that unadversarial learning 'can also be solved in a generative fashion.' The later claim that the VAE decoder represents FΦ and the encoder predicts ∂δ0/∂x is a relabeling of the residual, not a consequence of the iteration.

  2. other [Supplementary Sec. 8.2, Eqs. (23)-(30); used in Sec. 4.2]
    "we treat s as a middle variable, thus ∂δ0 ∂x can be expressed as the following equation by the chain law. ∂δ0 ∂x = ∂δ0 ∂s · ∂s ∂x ≤ U · ∂s ∂x , (23) ... The results presented above suggest a insight that ∂s ∂x is proportional to the saliency map s, namely ∂s ∂x ∝ s. (29)"

    Equation (10) defines δ0 as 'an initial random noise,' i.e., not as a function of x or of the saliency map s, so ∂δ0/∂x is zero or undefined and the chain rule ∂δ0/∂x = (∂δ0/∂s)(∂s/∂x) is not available. The proof supplies no argument that ∂δ0/∂s exists. Moreover, Eq. (27) gives ∂s/∂x = 1/2 − Σ IΔi/IΔ, a quantity depending on the local perturbation Δx, not on s; the proportionality ∂s/∂x ∝ s is asserted rather than derived. Therefore Theorem 2's upper bound, which is the paper's justification that the saliency map 'provides an upper bound' for the latent input, reduces to assuming the result it purports to prove.

1 more flagged steps
  1. self definitional [Sec. 4.2, 'Solution to problem A' and Eq. (4)]
    "In practice, considering derivative ∂δ0 ∂x is relevant with both random noise δ0 and input image x, we sample it from a certain Gaussian distribution associated with x. ... we approximate ∂δ0 ∂x by latent variable z, which is jointly determined by input x and sampled random signal τ."

    After Theorem 2 has failed to establish an independent meaning for ∂δ0/∂x, the paper simply declares that the VAE latent z 'approximates' that derivative. The training objective enforces only that z follows N(0,I) and that x+δ reproduces the saliency map; neither term measures any relationship to δ0 or to the unadversarial objective in Eq. (9). The identification of z with ∂δ0/∂x is therefore by fiat, and the claimed theoretical grounding of the generative reformulation is equivalent to the paper's own definition of the latent variable rather than a derived result.

full rationale

The circularity is internal to the paper's derivation chain, not a self-citation issue. Theorem 1 is made true by construction: after deriving only an upper bound, the proof defines FΦ as the bracketed residual and chooses V to force the equality, so the 'generative form' carries no information; the subsequent VAE instantiation merely names that residual. Theorem 2 needs ∂δ0/∂x, but δ0 is an initial random noise, so that derivative is zero or undefined, and the chain rule through the saliency map is unavailable; the proportionality ∂s/∂x ∝ s is asserted after an expression that depends on Δx, not on s. The method then identifies the VAE latent z with this undefined derivative and uses the saliency map as its alleged upper bound, which is precisely how the GUES objective is justified. Each load-bearing theoretical step reduces to a definition or assertion, so the central claim that GUES is grounded in a generative reformulation of unadversarial learning is forced by the paper's own construction. The experimental comparison also lacks a model-agnostic input-transformation baseline, but that is an experimental-control concern rather than circularity. No self-citation is load-bearing in this chain.

Assumptions & free parameters 4 free parameters · 4 assumptions · 2 invented entities

The paper introduces a new setting and a new architecture, but the theoretical load-bearing entities are invented quantities with no independent justification. The key parameters α, β and the VAE capacity are chosen by hand. The saliency map is a deterministic classical computation, which is a reasonable external anchor, but the claimed mathematical link between it and the latent variable is not established.

free parameters (4)
  • α (KL weight) = 1.0
    Set to 1.0 in Sec 5.2; sensitivity study shows small effect, but it is still a manually chosen constant.
  • β (MSE weight) = tuned in {0.0001, 0.01, 1}; sensitivity explored 0.00005 to 0.00014
    β is tuned to keep L_KL and L_MSE on the same scale; the exact final value is not reported in the paper.
  • VAE latent dimension = 10
    Chosen without analysis or ablation for the latent size.
  • Batch size, learning rate, optimizer = 64, 1e-5, SGD momentum 0.9
    Standard training choices, not justified in the paper.
assumptions (4)
  • ad hoc to paper ∂δ0/∂x is a well-defined quantity and can be sampled as a latent variable z from a Gaussian conditioned on x.
    Invoked in Sec 4.2; δ0 is random noise independent of x, so its derivative with respect to x is not well-defined in ordinary calculus.
  • ad hoc to paper The saliency map provides an upper bound on ∂δ0/∂x (Theorem 2).
    The proof in Sec 8.2 is invalid: the bound ∂δ0/∂s ≤ U is asserted without justification, and the claimed proportionality ∂s/∂x ∝ s is not established by the derived cases.
  • domain assumption The source model is frozen and cannot be differentiated, but can be queried at inference time.
    This is the core assumption of OMG-DA, stated in Sec 1 and used in all experiments.
  • domain assumption The source model is a DeiT-base trained on source data with cross-entropy and label smoothing.
    Specified in Sec 5.2 as the backbone for all experiments.
invented entities (2)
  • latent function input ∂δ0/∂x
    purpose: Claimed to be the input to the perturbation generation function F_Φ, modeled by the VAE encoder's latent z.
    No independent measurability or falsifiable defining relation; it is parameterized through the VAE, so the bound in Theorem 2 cannot be verified.
  • Generative unadversarial examples
    purpose: The decoder's output: images transformed to resemble their saliency maps, which are then fed to a frozen model.
    The term is new, but the examples are just saliency-like reconstructions; there is no falsifiable prediction outside the reported accuracy numbers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Domain Adaptive Diabetic Retinopathy Grading with Model Absence and Flowing Data." pith.science (2026). https://pith.science/paper/YZKKNBB2

@misc{pith2026241201203,
  author       = {Pith},
  title        = {Pith review of: Domain Adaptive Diabetic Retinopathy Grading with Model Absence and Flowing Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YZKKNBB2}},
  note         = {Machine review of arXiv:2412.01203}
}
read the original abstract

Domain shift (the difference between source and target domains) poses a significant challenge in clinical applications, e.g., Diabetic Retinopathy (DR) grading. Despite considering certain clinical requirements, like source data privacy, conventional transfer methods are predominantly model-centered and often struggle to prevent model-targeted attacks. In this paper, we address a challenging Online Model-aGnostic Domain Adaptation (OMG-DA) setting, driven by the demands of clinical environments. This setting is characterized by the absence of the model and the flow of target data. To tackle the new challenge, we propose a novel approach, Generative Unadversarial ExampleS (GUES), which enables adaptation from a data-centric perspective. Specifically, we first theoretically reformulate conventional perturbation optimization in a generative way--learning a perturbation generation function with a latent input variable. During model instantiation, we leverage a Variational AutoEncoder to express this function. The encoder with the reparameterization trick predicts the latent input, whilst the decoder is responsible for the generation. Furthermore, the saliency map is selected as pseudo-perturbation labels. Because it not only captures potential lesions but also theoretically provides an upper bound on the function input, enabling the identification of the latent variable. Extensive comparative experiments on DR benchmarks with both frozen pre-trained models and trainable models demonstrate the superiority of GUES, showing robustness even with small batch size.

Figures

Figures reproduced from arXiv: 2412.01203 by the authors.

Figure 1
Figure 1. Comparison between the OMG-DA and SFDA settings. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The instantiation framework of GUES in the OMG-DA setting. (a) For target input [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Explanation in choosing fine-grained saliency maps as [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Comparison results with batch size varying from 2 to 64 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Interpretability analysis based on a typical fundus image from “Moderate DR” class in APTOS. Here, H (hemorrhages), SE (soft [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Feature distribution comparison of 3D density charts on [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Parameter sensitiveness study results over [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Illustration of x + △x at coordinate (h, w) as we select the simplest surround case ς = 1. Thus, G(x + △x) at (h, w) can be expressed as Ghw(x + △x) = X ς max{  cen (h, w) − 1 8 sur (h, w, ς)  −  1 8 sur△(ς, △x) − I△  , 0} (26) Let A1 = cen (h, w) and A2 = sur (h, …
Figure 10
Figure 10. Figure 10: Visualize the styles and characteristics of each dataset by analyzing the RGB statistics of proliferative diabetic retinopathy (PDR) [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Visualization for input images, generative perturbations, and RGB statistic of the corresponding perturbations on transfer task [PITH_FULL_IMAGE:figures/full_fig_p014_11.png]
Figure 12
Figure 12. Figure 12: Visualization of a fundus image, a natural image, and their corresponding saliency maps. The fundus image is sampled from [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 38 canonical work pages

  1. [1]

    https: //www.kaggle.com/c/aptos2019- blindness- detection, accessed February 20, 2022

    Aptos: Aptos 2019 blindness detection website. https: //www.kaggle.com/c/aptos2019- blindness- detection, accessed February 20, 2022. 5, 13

    work page 2019

  2. [2]

    https://www.Eyepacs

    Quadratic weighted kappa. https://www.Eyepacs. com / aroraaman / quadratic - kappa - metric - explained - in - 5 - simple - steps, accessed July

    work page

  3. [3]

    A comprehensive diagnosis system for early signs and different diabetic retinopathy grades using fundus retinal images based on pathological changes detection

    Eman AbdelMaksoud, Sherif Barakat, and Mohammed El- mogy. A comprehensive diagnosis system for early signs and different diabetic retinopathy grades using fundus retinal images based on pathological changes detection. Computers in Biology and Medicine, 126:104039, 2020. 1

    work page 2020

  4. [4]

    Drgen: do- main generalization in diabetic retinopathy classification

    Mohammad Atwany and Mohammad Yaqub. Drgen: do- main generalization in diabetic retinopathy classification. In MICCAI, 2022. 2

    work page 2022

  5. [5]

    To- wards generalizable diabetic retinopathy grading in unseen domains

    Haoxuan Che, Yuhan Cheng, Haibo Jin, and Hao Chen. To- wards generalizable diabetic retinopathy grading in unseen domains. In MICCAI, 2023. 2

    work page 2023

  6. [6]

    A deep learning system for detecting diabetic retinopathy across the disease spectrum

    Ling Dai, Liang Wu, Huating Li, Chun Cai, Qiang Wu, Hongyu Kong, Ruhan Liu, Xiangning Wang, Xuhong Hou, Yuexing Liu, et al. A deep learning system for detecting diabetic retinopathy across the disease spectrum. Nature communications, 12(1):3242, 2021. 1

    work page 2021

  7. [7]

    Feedback on a publicly distributed image database: the mes- sidor database

    Etienne Decenci `ere, Xiwei Zhang, Guy Cazuguel, Bruno Lay, B´eatrice Cochener, Caroline Trone, Philippe Gain, John- Richard Ord´o˜nez-Varela, Pascale Massin, Ali Erginay, et al. Feedback on a publicly distributed image database: the mes- sidor database. Image Analysis & Stereology, pages 231–234,

  8. [8]

    Cabnet: Category attention block for imbalanced diabetic retinopathy grading

    Along He, Tao Li, Ning Li, Kai Wang, and Huazhu Fu. Cabnet: Category attention block for imbalanced diabetic retinopathy grading. IEEE Transactions on Medical Imaging, 40(1):143–153, 2020. 1

    work page 2020

Show all 46 references
  1. [9]

    Ssit: Saliency-guided self-supervised image transformer for diabetic retinopathy grading

    Yijin Huang, Junyan Lyu, Pujin Cheng, Roger Tam, and Xiaoying Tang. Ssit: Saliency-guided self-supervised image transformer for diabetic retinopathy grading. IEEE Journal of Biomedical and Health Informatics, 2024. 3

    2024

  2. [10]

    Auto-encoding variational bayes

    Diederik P Kingma. Auto-encoding variational bayes. arXiv:1312.6114, 2013. 2

    2013 arXiv

  3. [11]

    An introduction to domain adaptation and transfer learning

    Wouter M Kouw and Marco Loog. An introduction to domain adaptation and transfer learning. arXiv:1812.11806, 2018. 1

    2018 arXiv

  4. [12]

    Confidence score for source-free unsupervised domain adap- tation

    Jonghyun Lee, Dahuin Jung, Junho Yim, and Sungroh Yoon. Confidence score for source-free unsupervised domain adap- tation. In ICML, 2022. 6

    2022

  5. [13]

    Diagnostic assessment of deep learning algo- rithms for diabetic retinopathy screening

    Tao Li, Yingqi Gao, Kai Wang, Song Guo, Hanruo Liu, and Hong Kang. Diagnostic assessment of deep learning algo- rithms for diabetic retinopathy screening. Information Sci- ences, 501:511–522, 2019. 5, 13

    2019

  6. [14]

    Applications of deep learning in fundus images: A review

    Tao Li, Wang Bo, Chunyu Hu, Hong Kang, Hanruo Liu, Kai Wang, and Huazhu Fu. Applications of deep learning in fundus images: A review. Medical Image Analysis, 69: 101971, 2021. 1

    2021

  7. [15]

    Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation

    Jian Liang, Dapeng Hu, and Jiashi Feng. Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation. In ICML, 2020. 3, 5, 6, 14

    2020

  8. [16]

    Guid- ing pseudo-labels with uncertainty estimation for source-free unsupervised domain adaptation

    Mattia Litrico, Alessio Del Bue, and Pietro Morerio. Guid- ing pseudo-labels with uncertainty estimation for source-free unsupervised domain adaptation. In CVPR, 2023. 6

    2023

  9. [17]

    Deepdrid: Diabetic retinopathy—grading and image quality estimation challenge

    Ruhan Liu, Xiangning Wang, Qiang Wu, Ling Dai, Xi Fang, Tao Yan, Jaemin Son, Shiqi Tang, Jiang Li, Zijian Gao, et al. Deepdrid: Diabetic retinopathy—grading and image quality estimation challenge. Patterns, 3(6), 2022. 5, 13

    2022

  10. [18]

    Cat: Collaborative adversarial training

    Xingbin Liu, Huafeng Kuang, Xianming Lin, Yongjian Wu, and Rongrong Ji. Cat: Collaborative adversarial training. arXiv:2303.14922, 2023. 3

    2023 arXiv

  11. [19]

    Human detection using a mobile platform and novel features derived from a visual saliency mechanism

    Sebastian Montabone and Alvaro Soto. Human detection using a mobile platform and novel features derived from a visual saliency mechanism. Image and Vision Computing, 28 (3):391–402, 2010. 3

    2010

  12. [20]

    Self-supervised domain adap- tation for diabetic retinopathy grading using vessel image reconstruction

    Duy MH Nguyen, Truong TN Mai, Ngoc TT Than, Alexander Prange, and Daniel Sonntag. Self-supervised domain adap- tation for diabetic retinopathy grading using vessel image reconstruction. In KI, 2021. 2

    2021

  13. [21]

    Towards stable test-time adaptation in dynamic wild world

    Shuaicheng Niu, Jiaxiang Wu, Yifan Zhang, Zhiquan Wen, Yaofo Chen, Peilin Zhao, and Mingkui Tan. Towards stable test-time adaptation in dynamic wild world. In ICLR, 2023. 6

    2023

  14. [22]

    Augpaste: A one-shot approach for diabetic retinopathy detection

    Jiaming Qiu, Weikai Huang, Yijin Huang, Nanxi Yu, and Xiaoying Tang. Augpaste: A one-shot approach for diabetic retinopathy detection. Biomedical Signal Processing and Control, 96:106489, 2024. 3

    2024

  15. [23]

    Source-free active domain adaptation for diabetic retinopathy grading based on ultra-wide-field fundus images

    Jinye Ran, Guanghua Zhang, Fan Xia, Ximei Zhang, Juan Xie, and Hao Zhang. Source-free active domain adaptation for diabetic retinopathy grading based on ultra-wide-field fundus images. Computers in Biology and Medicine , 174:108418,

  16. [24]

    Unadversarial exam- ples: Designing objects for robust vision

    Hadi Salman, Andrew Ilyas, Logan Engstrom, Sai Vemprala, Aleksander Madry, and Ashish Kapoor. Unadversarial exam- ples: Designing objects for robust vision. In NeurIPS, 2021. 2, 3

    2021

  17. [25]

    Nsa: Naturalistic support artifact to boost network confidence

    Abhijith Sharma, Phil Munz, and Apurva Narayan. Nsa: Naturalistic support artifact to boost network confidence. In IJCNN, 2023. 3

    2023

  18. [26]

    Membership inference attacks against machine learning models

    Reza Shokri, Marco Stronati, Congzheng Song, and Vitaly Shmatikov. Membership inference attacks against machine learning models. In S&P, 2017. 1, 2

    2017

  19. [27]

    Screening for diabetic retinopathy

    Daniel E Singer, David M Nathan, Howard A Fogel, and An- drew P Schachat. Screening for diabetic retinopathy. Annals of Internal Medicine, 116(8):660–671, 1992. 1

    1992

  20. [28]

    Intriguing properties of neural networks

    C Szegedy. Intriguing properties of neural networks. arXiv:1312.6199, 2013. 3 9

    2013 arXiv

  21. [29]

    Source-free domain adaptation via target prediction distribution searching

    Song Tang, An Chang, Fabian Zhang, Xiatian Zhu, Mao Ye, and Changshui Zhang. Source-free domain adaptation via target prediction distribution searching. International Journal of Computer Vision, 132(3):654–672, 2024. 3, 6

    2024

  22. [30]

    Unified source-free domain adaptation

    Song Tang, Wenxin Su, Mao Ye, Jianwei Zhang, and Xiatian Zhu. Unified source-free domain adaptation. arXiv:2403.07601, 2024. 3

    2024

  23. [31]

    A visual attention-based algorithm for brain tumor detec- tion using an on-center saliency map and a superpixel-based framework

    Nishtha Tomar, Sushmita Chandel, and Gaurav Bhatnagar. A visual attention-based algorithm for brain tumor detec- tion using an on-center saliency map and a superpixel-based framework. Healthcare Analytics, 5:100323, 2024. 3

    2024

  24. [32]

    Training data-efficient image transformers & distillation through atten- tion

    Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Herv´e J´egou. Training data-efficient image transformers & distillation through atten- tion. In ICML, 2021. 5

    2021

  25. [33]

    On-the-fly test-time adaptation for medical image segmentation

    Jeya Maria Jose Valanarasu, Pengfei Guo, VS Vibashan, and Vishal M Patel. On-the-fly test-time adaptation for medical image segmentation. In MIDL, 2024. 2

    2024

  26. [34]

    Deep hashing network for unsupervised domain adaptation

    Hemanth Venkateswara, Jose Eusebio, Shayok Chakraborty, and Sethuraman Panchanathan. Deep hashing network for unsupervised domain adaptation. In CVPR, 2017. 14

    2017

  27. [35]

    Tent: Fully test-time adaptation by entropy minimization

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Ol- shausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization. In ICLR, 2020. 2, 3, 6, 14

    2020

  28. [36]

    Saliency-guided and patch-based mixup for long-tailed skin cancer image classification

    Tianyunxi Wei, Yijin Huang, Li Lin, Pujin Cheng, Sirui Li, and Xiaoying Tang. Saliency-guided and patch-based mixup for long-tailed skin cancer image classification. arXiv:2406.10801, 2024. 3

    2024 arXiv

  29. [37]

    Coarse-to-fine classification for diabetic retinopathy grading using convolutional neural network

    Zhan Wu, Gonglei Shi, Yang Chen, Fei Shi, Xinjian Chen, Gouenou Coatrieux, Jian Yang, Limin Luo, and Shuo Li. Coarse-to-fine classification for diabetic retinopathy grading using convolutional neural network. Artificial Intelligence in Medicine, 108:101936, 2020. 1

    2020

  30. [38]

    Cdtrans: Cross-domain transformer for unsupervised domain adaptation

    Tongkun Xu, Weihua Chen, W ANG Pichao, Fan Wang, Hao Li, and Rong Jin. Cdtrans: Cross-domain transformer for unsupervised domain adaptation. In ICLR, 2021. 5

    2021

  31. [39]

    Exploiting the intrinsic neighborhood structure for source-free domain adaptation

    Shiqi Yang, Joost van de Weijer, Luis Herranz, Shangling Jui, et al. Exploiting the intrinsic neighborhood structure for source-free domain adaptation. In NeurIPS, 2021. 3, 5, 6

    2021

  32. [40]

    See through gradients: Image batch recovery via gradinversion

    Hongxu Yin, Arun Mallya, Arash Vahdat, Jose M Alvarez, Jan Kautz, and Pavlo Molchanov. See through gradients: Image batch recovery via gradinversion. In CVPR, 2021. 2

    2021

  33. [41]

    Diabetic retinopathy grading by a source-free transfer learning approach

    Chenrui Zhang, Tao Lei, and Ping Chen. Diabetic retinopathy grading by a source-free transfer learning approach. Biomedi- cal Signal Processing and Control, 73:103423, 2022. 2 10 Domain Adaptive Diabetic Retinopathy Grading with Model Absence and Flowing Data Supplementary Material

    2022

  34. [42]

    Reproducibility Statement The code and data will be made available after the publica- tion of this paper

  35. [43]

    A Proof of Theorem 1 Recalling traditional unadversarial learning

    Proof of Theorem 8.1. A Proof of Theorem 1 Recalling traditional unadversarial learning. Unadver- sarial learning aims to develop an image perturbation that enhances the performance on a specific class, which can be succinctly described as follows: ˆδ = arg min δ L(fθ(x + δ), ...

  36. [44]

    No DR” class comprises about 49.2% of all samples. In DDR, “No DR

    Implementation Details 9.1. Datasets Details Dataset description. We evaluate the proposed method on four standard DR benchmarks. Their details are presented as follows. 12 Dataset Statistics APTOS, India Dataset Statistics DDR, China Dataset Statistics DeepDR, China PDRDatase...

  37. [45]

    The computation rules for accuracy (termed ACC), Quadratic Weighted Kappa (termed QWK), and the average of QWK and ACC (termed A VG) are as follows

    Evaluation metrics. The computation rules for accuracy (termed ACC), Quadratic Weighted Kappa (termed QWK), and the average of QWK and ACC (termed A VG) are as follows. ACC = T P+ T N T P+ T N+ F P+ F N, QW K= 1− Pn i=1 Pn j=1 W (i, j) · O(i, j)Pn i=1 Pn j=1 W (i, j) · E(i, j)...

  38. [46]

    Results with Varying Batch Size As a supplement to the results with varying batch sizes, Ta- ble 5 presents the complete performance of three evaluation metrics across all 12 tasks

    Supplementary Experiment Results 11.1. Results with Varying Batch Size As a supplement to the results with varying batch sizes, Ta- ble 5 presents the complete performance of three evaluation metrics across all 12 tasks. TTA methods SHOT-IM and TENT show a performance drop whe...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.