A Super-pixel-based Approach to the Stable Interpretation of Neural Networks

Farzan Farnia; Jingwei Zhang; Qi Dou; Shizhan Gong

REVIEW 2 major objections 4 minor 41 references

A Super-pixel-based Approach to the Stable Interpretation of Neural Networks

T0 review · 2 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Averaging gradient-based saliency scores within super-pixels reduces the run-to-run variance and generalization error of neural-network explanations, with little or no loss in fidelity.

desk verdict Useful empirical method, but the main theoretical claim about reducing generalization error does not follow from the paper's own stability bound. read the letter →

arxiv 2412.14509 v1 pith:QSWTK5IY submitted 2024-12-19 cs.CV

Shizhan Gong , Jingwei Zhang , Qi Dou , Farzan Farnia This is my paper · ORCID

classification cs.CV

keywords saliencymapsuper-pixelinterpretationstabilityalgorithmicgeneralizationerrorgradient-basedmodelinterpretability

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

Saliency maps produced by gradient-based interpretability methods change noticeably when a neural network is retrained on different data or with different random seeds, making them unreliable for understanding what a model is actually looking at. This paper argues that much of that instability can be removed by a simple preprocessing step: partition the image into super-pixels (perceptually homogeneous regions) and replace each pixel's importance score with the average score over its super-pixel. The paper proves that this grouping cannot increase the interpretation generalization error, defined as the expected gap between the saliency map computed on finite training data and the saliency map of the population-optimal network, and it demonstrates on CIFAR-10 and ImageNet that super-pixel maps are more stable across networks, more reproducible under data removal, and equally or more interpretable, with only small and controllable losses in fidelity. The reason the trick works is that averaging inside each group suppresses the high-dimensional gradient noise that stochastic training introduces, while the semantic alignment of super-pixels preserves the signal a human would want the explanation to carry.

What carries the argument

The central object is the grouped-gradient operator g-SG( f, x) = κ(x)∇x f (x), where κ(x) is a d×d matrix that replaces the gradient at each pixel by the mean gradient over its super-pixel (κ(x)_{i,j} = 1/|S_k| when pixels i and j lie in the same group S_k, and 0 otherwise). This operator carries the argument: it is the device that converts a pixel-level explanation into a region-level one without retraining the network. The theoretical machinery is the algorithmic-stability framework: Theorem 1 converts ε-uniform stability of an interpretation scheme into a bound on its interpretation generalization error, Proposition 3 shows grouping preserves or improves the stability constant, and Proposition 2 quantifies the variance that the averaging removes. The super-pixel algorithms (SLIC, Quickshift, graph-based methods) supply the partition; they matter because the variance-reduction benefit is exactly the within-group variance of the gradient difference, which is small when groups are semantically homogeneous.

What would settle it

Train one architecture on two disjoint halves of a dataset and, for images where thin or high-frequency structures deliberately cross super-pixel boundaries, compute both pixel and super-pixel saliency maps. If the super-pixel map's MeGe generalization error does not fall below the pixel map's, or if its deletion curve removes little predicted class probability until regions far from the true object are erased, the within-group homogeneity assumption is violated and the variance-reduction claim fails for that partition.

Watch

Extended reading notes

Core claim

The central claim is that for any neural network classifier, replacing the pixel-wise Simple Gradient map with a grouped map built from super-pixels reduces the expected deviation of the empirical explanation from the population explanation. Formally, the paper defines the interpretation loss as the L2 distance between the empirical map and the map of the population-trained network, and the interpretation generalization error as the expected gap between that loss on test data and on training data. It proves, using Jensen's inequality, that the grouped loss is pointwise no larger than the pixel-wise loss (Proposition 1), and that the difference equals the sum, over super-pixels, of the super-pixel size times the within-group variance of the gradient difference (Proposition 2). Extending the algorithmic-stability framework to explanations, it shows that if the pixel-wise map is ε-uniformly stable, the grouped map is stable with a no-worse constant (Proposition 3), and therefore εgen(g-SG) ≤ εgen(SG) (Corollary 1). The paper reads this as variance reduction by averaging: grouping lowers the effective dimension of the input, and the noise that stochastic training leaves in the gradient is averaged out within semantically meaningful regions.

Load-bearing premise

The argument stands on the assumption that every pixel inside a super-pixel has the same true importance to the network's decision; if a super-pixel crosses an object boundary or contains mixed textures, averaging will blur the very detail the explanation is meant to show.

Editorial extensions

If this is right

Any gradient-based saliency method (Simple Gradient, Integrated Gradients, SmoothGrad, Sparsified SmoothGrad) can be made more reproducible across training runs by averaging the map within super-pixels, at the cost of a tunable and usually small fidelity loss.
Small super-pixels (about 1000–2000 groups on ImageNet) give most of the stability gain with negligible or even positive fidelity changes; overly large super-pixels degrade both stability and fidelity.
The grouping strategy is orthogonal to SmoothGrad-style noise averaging, so the two mechanisms can be combined to push stability further.
Because the partition is static and semantic, the method is computationally cheap relative to region-based approaches such as RISE, XRAI, and Score-CAM, and it can be applied to any gradient-based interpretation scheme.

Reading between the lines

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

The same averaging argument should apply to any high-dimensional input with a known grouping structure — audio frequency bands, gene modules, or time windows — as long as the groups align with the true decision-relevant features, suggesting a generic recipe for stabilizing explanations beyond images.
The paper's theory predicts the stability gain is exactly the within-group variance of the gradient difference (Proposition 2). A direct test of the mechanism would be to measure that variance at object boundaries and verify that the SSIM improvement tracks its magnitude; if the improvement persists even where the variance is high, the explanation would need to be revised.
An adaptive partition — small super-pixels near boundaries and large ones in uniform regions — is a natural next step that could push the fidelity–stability frontier beyond the fixed-size partitions used here.

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.

Desk Editor's Note

read the letter

Short version: the method works empirically, and the stability framing is a reasonable way to think about the problem, but the paper's headline theoretical claim—Corollary 1—doesn't follow from their own results. That gap is real and load-bearing, because the corollary is what the title promises.

The idea is simple and useful: partition the image into super-pixels, average the gradient-based saliency within each group, and you get maps that are more stable across training runs. The experiments on CIFAR-10 and ImageNet back this up across several gradient methods, and the fidelity trade-off is modest. The comparison against square patches and random partitions is a nice control. Code is released. That is a solid empirical contribution.

The theory is where it falls down. Propositions 1 and 2 are fine applications of Jensen and variance decomposition. Proposition 3 presumably holds, since averaging with an orthogonal projection is non-expansive. But Corollary 1 is not a consequence of Proposition 3 plus Theorem 1. Theorem 1 gives an upper bound on generalization error in terms of the stability constant. A smaller stability constant for g-SG gives a smaller upper bound, not a smaller actual error. To conclude ε_gen(g-SG) ≤ ε_gen(SG), you'd need either a lower bound on SG's error or a direct bias–variance argument. Neither is in the main text, and the appendix is not in this preprint. The pointwise domination in Proposition 1 doesn't save it either, because the absolute gap between population and empirical means can behave differently. So the central theoretical claim is unsupported as written.

There is also the load-bearing assumption that pixels within a super-pixel share the same true importance. That is fine when super-pixels respect object boundaries, but it will blur real detail when they do not. The empirically chosen sizes (1000–2000 groups) seem to keep this in check, but it is a limitation worth stating more sharply.

Who is this for? Anyone working on interpretability stability or applying saliency maps to high-stakes domains. The method is cheap, easy to implement, and likely to be useful even if the theory is reframed as “the worst-case bound improves” rather than “the generalization error decreases.”

Recommendation: send to peer review. A good referee can catch the corollary issue, and the empirical contribution plus the correct propositions are worth publishing. I would suggest the authors either prove a lower bound for SG's error or explicitly reframe the claim as a stability improvement with a tighter bound.

Referee Report

2 major / 4 minor

Summary. The paper proposes a super-pixel-based grouping strategy to stabilize gradient-based saliency maps. The authors define an interpretation generalization error, prove a stability-based bound (Theorem 1), and claim that grouping pixels into super-pixels reduces this error (Corollary 1). They support the claim with Propositions 1–3 on the variance-reducing effect of averaging within groups, and they report experiments on CIFAR-10 and ImageNet measuring SSIM, MeGe, fidelity, and interpretability, which show consistent improvements over pixel-based maps.

Significance. If the theoretical claim were valid, the paper would contribute a simple, computationally inexpensive, and broadly applicable method for producing more stable and generalizable saliency maps, backed by a formal guarantee. The empirical study is extensive, includes multiple interpretation methods and metrics, and makes the code available. The averaging/grouping insight is natural and the propositions on variance reduction via Jensen's inequality are correct. However, the central theoretical corollary does not follow from the stated results, and the proof of the main theorem is not included in the preprint. As written, the paper establishes an improvement in the stability bound, not a reduction of the actual generalization error, which materially weakens the theoretical contribution.

major comments (2)

[Section 3.2, Corollary 1] The deduction 'εgen(g-SG) ≤ εgen(SG)' does not follow from Proposition 3 and Theorem 1. Proposition 3 states that if SG is ε-uniformly stable, then g-SG is ε'-uniformly stable for some ε' ≤ ε. Theorem 1 gives εgen(SG) ≤ ε and εgen(g-SG) ≤ ε'. These inequalities imply that both generalization errors are at most ε, with g-SG having a tighter upper bound, but they do not imply that the actual value of εgen(g-SG) is no larger than the actual value of εgen(SG). This is the fallacy of bounding a bound. To establish the corollary, the authors would need either a lower bound on εgen(SG) in terms of its stability constant or a direct comparison of the two generalization errors, for example via the bias-variance decomposition expressed in Proposition 2. Since Corollary 1 is the stated theoretical basis for the central claim that 'the grouping of pixels reduces the generalization error,' this is a load-bearing gap.
[Section 3.1, Theorem 1] Theorem 1 is stated as a general bound 'εgen(I) ≤ ε' without specifying the conditions under which this stability-based generalization bound holds. In the standard algorithmic-stability framework, such bounds require assumptions such as symmetry of the algorithm and, for the absolute-value form used in Definition 2, careful handling of the expectation. The proof is deferred to an appendix that is not included in this preprint, so the correctness of the theorem cannot be verified from the manuscript. Because Theorem 1 is the foundation for the corollary, the authors should either provide the proof in the main text or appendix and state the precise assumptions (e.g., boundedness of the interpretation loss, if needed).

minor comments (4)

[Section 3.2, Definition of g-SG] The formula 'g-SG( f̂θ ,x) := AT ∇g f̂θ (x + A Wg)' appears to have a dimension error: A is d×p, so A^T is p×d, and A^T ∇g is not a well-defined d-dimensional vector. The subsequent simplification to κ(x)∇x f suggests that the intended broadcasting operator is A, not A^T (i.e., g-SG = A ∇g f = A W A^T ∇x f). Please correct this definition.
[Section 3.2, Notation] The sentence 'We let g = 0p to be the p-dimensional zero vector' is confusing because g is later used as a variable in the gradient. It would be clearer to say that g is a p-dimensional group-level input variable, and the gradient is evaluated at g = 0.
[Section 3.3, Super-pixel assumption] The assumption that pixels within a super-pixel have essentially the same effect on the prediction is stated qualitatively and is crucial for the fidelity of the resulting maps. The main text defers validation to the appendix; a direct measurement of within-super-pixel gradient variance, or a comparison of grouped maps against ground-truth object boundaries, would strengthen the justification.
[Figure 2 and Table 1] The abbreviations 'MeGe', 'pixel', 'super', 'sup.', and 'rand' are used in figures and tables without full expansion in the captions; please define them for the reader.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the super-pixel averaging derivation is self-contained; the flawed Corollary 1 is a logical gap, not a circular argument.

full rationale

The derivation is self-contained and does not reduce to its own inputs. g-SG is defined as g-SG(fθ,x) = κ(x)∇x fθ(x), where κ is a group-averaging (orthogonal projection) operator; Proposition 1 follows from Jensen's inequality applied to the norm-squared loss, and Proposition 2 is an explicit variance identity for the loss gap. Proposition 3, assuming the deferred appendix supplies the standard non-expansiveness argument for averaging, is also a direct consequence of the definitions. No fitted parameter is renamed as a prediction: the super-pixel algorithm and its size are external, unsupervised choices, and the theoretical claims do not depend on their tuned values. The self-citations [10], [11], [12], [40] are contextual and not load-bearing; the paper's main evaluation uses external metrics (SSIM between independently trained models, MeGe, ROAR/ROAD, deletion/insertion, and µFidelity). The stated 'assumption that pixels inside a super-pixel have the same effect' is an explicit modeling assumption, not a hidden restatement of the result. The one serious issue is Corollary 1: 'If SG on a neural network f is ε-uniformly stable, then its generalization error can be reduced by the grouping of pixels, i.e., εgen(g-SG) ≤ εgen(SG).' Combining Theorem 1 (εgen(I) ≤ ε) with Proposition 3 (ε′ ≤ ε) does not imply εgen(g-SG) ≤ εgen(SG), because both quantities are only shown to share upper bounds and the upper bound on SG is not a lower bound on its actual generalization error. That is a logical gap, not a circular derivation, so it does not raise the circularity score.

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

The central claim depends on a small set of domain assumptions about super-pixel semantic coherence and the existence of a population saliency map. There are no invented physical entities. The main free parameters are the super-pixel granularity and algorithm, which are dataset choices rather than fitted constants.

free parameters (2)

number of super-pixels / max distance = 100 SLIC segments (CIFAR-10); Quickshift max distance 3 (ImageNet)
These hyperparameters control group size and are chosen per dataset. Section 4.2 reports minimal sensitivity, but they are still user-set and affect the stability-fidelity tradeoff.
super-pixel algorithm = SLIC on CIFAR-10, Quickshift on ImageNet
The choice of segmentation algorithm is a free choice; the paper shows robustness to it but uses different algorithms for different datasets.

assumptions (3)

domain assumption Pixels within a super-pixel have nearly identical importance for the classifier's decision.
Section 3.3 states 'the assumption that pixels inside a super-pixel have the same effect on the model's prediction and interpretation sounds relevant.' The whole interpretability benefit rests on this; otherwise averaging destroys information.
domain assumption A unique population saliency map exists for the trained classifier.
Section 3.1: 'the definition of the interpretation loss based on the assumption that there is a unique population saliency map, which is generally true for classification networks trained with cross-entropy according to the directional convergence theory [16].' This justifies measuring the deviation from a fixed target.
standard math The Bousquet-Elisseeff algorithmic stability framework applies to interpretation algorithms with the given loss.
Theorem 1 is an application of stability-based generalization bounds; the proof is in the Appendix. The loss L is a norm distance, and the bound may require additional assumptions (e.g., boundedness) not stated in the main text.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Super-pixel-based Approach to the Stable Interpretation of Neural Networks." pith.science (2026). https://pith.science/paper/QSWTK5IY

@misc{pith2026241214509,
  author       = {Pith},
  title        = {Pith review of: A Super-pixel-based Approach to the Stable Interpretation of Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QSWTK5IY}},
  note         = {Machine review of arXiv:2412.14509}
}

read the original abstract

Saliency maps are widely used in the computer vision community for interpreting neural network classifiers. However, due to the randomness of training samples and optimization algorithms, the resulting saliency maps suffer from a significant level of stochasticity, making it difficult for domain experts to capture the intrinsic factors that influence the neural network's decision. In this work, we propose a novel pixel partitioning strategy to boost the stability and generalizability of gradient-based saliency maps. Through both theoretical analysis and numerical experiments, we demonstrate that the grouping of pixels reduces the variance of the saliency map and improves the generalization behavior of the interpretation method. Furthermore, we propose a sensible grouping strategy based on super-pixels which cluster pixels into groups that align well with the semantic meaning of the images. We perform several numerical experiments on CIFAR-10 and ImageNet. Our empirical results suggest that the super-pixel-based interpretation maps consistently improve the stability and quality over the pixel-based saliency maps.

Figures

Figures reproduced from arXiv: 2412.14509 by the authors.

**Figure 2.** Left: The SSIM of saliency map between two models trained with disjoint training datasets or different initialization from CIFAR10 (top) and Imagenet (bottom). Middle: Comparison of SSIM with region-based interpretation. Right: Comparison of MeGe between pixel-based and super-pixel-based methods. L2-norm is used as the distance measure. could be nearly independent, suggesting that applying super-pixels for groupin… view at source ↗

**Figure 3.** Qualitative comparison on ImageNet dataset between pixel-based and super-pixel [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Visualization of super-pixel-based simple gradient maps with different numbers of [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]

**Figure 5.** Figure 5: Left: Tradeoff curve between SSIM and µFidelity. The size of the dots reflects the size of super-pixels. Middle: Comparison of ROAD (top) and ROAR (bottom) between pixel-based and super-pixel-based methods. Right: L2-norm difference between super-pixelbased SimpleGrad…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 27 canonical work pages

[1]

Slic superpixels compared to state-of-the-art superpixel methods

Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua, and Sabine Süsstrunk. Slic superpixels compared to state-of-the-art superpixel methods. IEEE transactions on pattern analysis and machine intelligence , 34(11):2274–2282, 2012

work page 2012
[2]

Assessing the trustworthiness of saliency maps for localizing abnormalities in medical imaging

Nishanth Arun, Nathan Gaw, Praveer Singh, Ken Chang, Mehak Aggarwal, Bryan Chen, Katharina Hoebel, Sharut Gupta, Jay Patel, Mishka Gidwani, et al. Assessing the trustworthiness of saliency maps for localizing abnormalities in medical imaging. Radiology: Artificial Intelligence, 3(6):e200267, 2021

work page 2021
[3]

Evaluating and aggregating feature- based model explanations

Umang Bhatt, Adrian Weller, and José MF Moura. Evaluating and aggregating feature- based model explanations. arXiv preprint arXiv:2005.00631, 2020

arXiv 2005
[4]

Deep- learning-assisted diagnosis for knee magnetic resonance imaging: development and retrospective validation of mrnet

Nicholas Bien, Pranav Rajpurkar, Robyn L Ball, Jeremy Irvin, Allison Park, Erik Jones, Michael Bereket, Bhavik N Patel, Kristen W Yeom, Katie Shpanskaya, et al. Deep- learning-assisted diagnosis for knee magnetic resonance imaging: development and retrospective validation of mrnet. PLoS medicine, 15(11):e1002699, 2018

work page 2018
[5]

Stability and generalization

Olivier Bousquet and André Elisseeff. Stability and generalization. The Journal of Machine Learning Research, 2:499–526, 2002

work page 2002
[6]

Grad-cam++: Generalized gradient-based visual explanations for deep convolu- tional networks

Aditya Chattopadhay, Anirban Sarkar, Prantik Howlader, and Vineeth N Balasubrama- nian. Grad-cam++: Generalized gradient-based visual explanations for deep convolu- tional networks. In 2018 IEEE winter conference on applications of computer vision (WACV), pages 839–847. IEEE, 2018

work page 2018
[7]

How good is your ex- planation? algorithmic stability measures to assess the quality of explanations for deep neural networks

Thomas Fel, David Vigouroux, Rémi Cadène, and Thomas Serre. How good is your ex- planation? algorithmic stability measures to assess the quality of explanations for deep neural networks. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 720–730, 2022

work page 2022
[8]

Efficient belief propagation for early vision

Pedro F Felzenszwalb and Daniel P Huttenlocher. Efficient belief propagation for early vision. International journal of computer vision, 70:41–54, 2006

work page 2006

Show all 41 references

[9]

Sci- entific inference with interpretable machine learning: Analyzing models to learn about real-world phenomena

Timo Freiesleben, Gunnar König, Christoph Molnar, and Alvaro Tejero-Cantero. Sci- entific inference with interpretable machine learning: Analyzing models to learn about real-world phenomena. arXiv preprint arXiv:2206.05487, 2022

2022 arXiv
[10]

Diffusion model based semi-supervised learning on brain hemorrhage images for efficient midline shift quantification

Shizhan Gong, Cheng Chen, Yuqi Gong, Nga Yan Chan, Wenao Ma, Calvin Hoi-Kwan Mak, Jill Abrigo, and Qi Dou. Diffusion model based semi-supervised learning on brain hemorrhage images for efficient midline shift quantification. In International Conference on Information Processin...

2023
[11]

3dsam-adapter: Holistic adaptation of sam from 2d to 3d for promptable medical image segmentation

Shizhan Gong, Yuan Zhong, Wenao Ma, Jinpeng Li, Zhao Wang, Jingyang Zhang, Pheng-Ann Heng, and Qi Dou. 3dsam-adapter: Holistic adaptation of sam from 2d to 3d for promptable medical image segmentation. arXiv preprint arXiv:2306.13465, 2023

2023 arXiv
[12]

Structured gradient-based interpretations via norm-regularized adversarial training

Shizhan Gong, Qi Dou, and Farzan Farnia. Structured gradient-based interpretations via norm-regularized adversarial training. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11009–11018, 2024

2024
[13]

Collection-cam: A faster region-based saliency method using collection-wise mask over pyramidal features.IEEE Access, 10:112776– 112788, 2022

Yungi Ha and Chan-Hyun Youn. Collection-cam: A faster region-based saliency method using collection-wise mask over pyramidal features.IEEE Access, 10:112776– 112788, 2022

2022
[14]

Deep residual learning for image recognition

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

2016
[15]

Evaluating fea- ture importance estimates

Sara Hooker, Dumitru Erhan, Pieter-Jan Kindermans, and Been Kim. Evaluating fea- ture importance estimates. arXiv preprint arXiv:1806.10758, 2, 2018

2018 arXiv
[16]

Directional convergence and alignment in deep learning

Ziwei Ji and Matus Telgarsky. Directional convergence and alignment in deep learning. Advances in Neural Information Processing Systems, 33:17176–17186, 2020

2020
[17]

Transdose: Transformer-based radiotherapy dose prediction from ct images guided by super-pixel-level gcn classification

Zhengyang Jiao, Xingchen Peng, Yan Wang, Jianghong Xiao, Dong Nie, Xi Wu, Xin Wang, Jiliu Zhou, and Dinggang Shen. Transdose: Transformer-based radiotherapy dose prediction from ct images guided by super-pixel-level gcn classification. Medical Image Analysis, page 102902, 2023

2023
[18]

Xrai: Bet- ter attributions through regions

Andrei Kapishnikov, Tolga Bolukbasi, Fernanda Viégas, and Michael Terry. Xrai: Bet- ter attributions through regions. In Proceedings of the IEEE/CVF International Con- ference on Computer Vision, pages 4948–4957, 2019

2019
[19]

Why are saliency maps noisy? cause of and solution to noisy saliency maps

Beomsu Kim, Junghoon Seo, Seunghyeon Jeon, Jamyoung Koo, Jeongyeol Choe, and Taegyun Jeon. Why are saliency maps noisy? cause of and solution to noisy saliency maps. In 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), pages 4149–4157. IEEE, 2019

2019
[20]

Interpretable learning for self-driving cars by visualizing causal attention

Jinkyu Kim and John Canny. Interpretable learning for self-driving cars by visualizing causal attention. In Proceedings of the IEEE international conference on computer vision, pages 2942–2950, 2017

2017
[21]

Imagenet classification with deep convolutional neural networks

Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing sys- tems, 25, 2012

2012
[22]

Certifiably robust interpretation in deep learning

Alexander Levine, Sahil Singla, and Soheil Feizi. Certifiably robust interpretation in deep learning. arXiv preprint arXiv:1905.12105, 2019

1905 arXiv
[23]

Super- pixel guided low-light images enhancement with features restoration

Xiaoming Liu, Yan Yang, Yuanhong Zhong, Dong Xiong, and Zhiyong Huang. Super- pixel guided low-light images enhancement with features restoration. Sensors, 22(10): 3667, 2022. GONG ET AL: SUPER-PIXEL-BASED INTERPRETA TION 13

2022
[24]

Detection of anaemia from retinal fundus images via deep learning.Nature Biomedical Engineering, 4(1):18–27, 2020

Akinori Mitani, Abigail Huang, Subhashini Venugopalan, Greg S Corrado, Lily Peng, Dale R Webster, Naama Hammel, Yun Liu, and Avinash V Varadarajan. Detection of anaemia from retinal fundus images via deep learning.Nature Biomedical Engineering, 4(1):18–27, 2020

2020
[25]

Compact watershed and preemptive slic: On improving trade-offs of superpixel segmentation algorithms

Peer Neubert and Peter Protzel. Compact watershed and preemptive slic: On improving trade-offs of superpixel segmentation algorithms. In 2014 22nd international confer- ence on pattern recognition, pages 996–1001. IEEE, 2014

2014
[26]

Rise: Randomized input sampling for explanation of black-box models

Vitali Petsiuk, Abir Das, and Kate Saenko. Rise: Randomized input sampling for explanation of black-box models. arXiv preprint arXiv:1806.07421, 2018

2018 arXiv
[27]

A consistent and efficient evaluation strategy for attribution methods

Yao Rong, Tobias Leemann, Vadim Borisov, Gjergji Kasneci, and Enkelejda Kasneci. A consistent and efficient evaluation strategy for attribution methods. arXiv preprint arXiv:2202.00449, 2022

2022 arXiv
[28]

Grad-cam: Visual explanations from deep networks via gradient-based localization

Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE international conference on computer vision, pages 618–626, 2017

2017
[29]

Learning important features through propagating activation differences

Avanti Shrikumar, Peyton Greenside, and Anshul Kundaje. Learning important features through propagating activation differences. In International conference on machine learning, pages 3145–3153. PMLR, 2017

2017
[30]

Deep inside convolutional networks: Visualising image classification models and saliency maps

Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013

2013 arXiv
[31]

Smoothgrad: removing noise by adding noise

Daniel Smilkov, Nikhil Thorat, Been Kim, Fernanda Viégas, and Martin Wattenberg. Smoothgrad: removing noise by adding noise. arXiv preprint arXiv:1706.03825, 2017

2017 arXiv
[32]

Striving for simplicity: The all convolutional net

Jost Tobias Springenberg, Alexey Dosovitskiy, Thomas Brox, and Martin Riedmiller. Striving for simplicity: The all convolutional net. arXiv preprint arXiv:1412.6806 , 2014

2014 arXiv
[33]

Axiomatic attribution for deep net- works

Mukund Sundararajan, Ankur Taly, and Qiqi Yan. Axiomatic attribution for deep net- works. In International conference on machine learning , pages 3319–3328. PMLR, 2017

2017
[34]

Efficientnet: Rethinking model scaling for convolutional neural networks

Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning, pages 6105–6114. PMLR, 2019

2019
[35]

Quick shift and kernel methods for mode seek- ing

Andrea Vedaldi and Stefano Soatto. Quick shift and kernel methods for mode seek- ing. In Computer Vision–ECCV 2008: 10th European Conference on Computer Vision, Marseille, France, October 12-18, 2008, Proceedings, Part IV 10 , pages 705–718. Springer, 2008

2008
[36]

Dynamic super-pixel normal- ization for robust hyperspectral image classification.IEEE Transactions on Geoscience and Remote Sensing, 61:1–13, 2023

Cong Wang, Lei Zhang, Wei Wei, and Yanning Zhang. Dynamic super-pixel normal- ization for robust hyperspectral image classification.IEEE Transactions on Geoscience and Remote Sensing, 61:1–13, 2023. 14 GONG ET AL: SUPER-PIXEL-BASED INTERPRETA TION

2023
[37]

Score-cam: Score-weighted visual explanations for convo- lutional neural networks

Haofan Wang, Zifan Wang, Mengnan Du, Fan Yang, Zijian Zhang, Sirui Ding, Piotr Mardziel, and Xia Hu. Score-cam: Score-weighted visual explanations for convo- lutional neural networks. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition workshop...

2020
[38]

Initialization noise in image gradients and saliency maps

Ann-Christin Woerl, Jan Disselhoff, and Michael Wand. Initialization noise in image gradients and saliency maps. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1766–1775, 2023

2023
[39]

Visualizing and understanding convolutional net- works

Matthew D Zeiler and Rob Fergus. Visualizing and understanding convolutional net- works. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzer- land, September 6-12, 2014, Proceedings, Part I 13, pages 818–833. Springer, 2014

2014
[40]

Moreaugrad: Sparse and robust interpretation of neural networks via moreau envelope

Jingwei Zhang and Farzan Farnia. Moreaugrad: Sparse and robust interpretation of neural networks via moreau envelope. arXiv preprint arXiv:2302.05294, 2023

2023 arXiv
[41]

Object detection with deep learning: A review

Zhong-Qiu Zhao, Peng Zheng, Shou-tao Xu, and Xindong Wu. Object detection with deep learning: A review. IEEE transactions on neural networks and learning systems, 30(11):3212–3232, 2019

2019

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Slic superpixels compared to state-of-the-art superpixel methods

Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua, and Sabine Süsstrunk. Slic superpixels compared to state-of-the-art superpixel methods. IEEE transactions on pattern analysis and machine intelligence , 34(11):2274–2282, 2012

work page 2012

[2] [2]

Assessing the trustworthiness of saliency maps for localizing abnormalities in medical imaging

Nishanth Arun, Nathan Gaw, Praveer Singh, Ken Chang, Mehak Aggarwal, Bryan Chen, Katharina Hoebel, Sharut Gupta, Jay Patel, Mishka Gidwani, et al. Assessing the trustworthiness of saliency maps for localizing abnormalities in medical imaging. Radiology: Artificial Intelligence, 3(6):e200267, 2021

work page 2021

[3] [3]

Evaluating and aggregating feature- based model explanations

Umang Bhatt, Adrian Weller, and José MF Moura. Evaluating and aggregating feature- based model explanations. arXiv preprint arXiv:2005.00631, 2020

arXiv 2005

[4] [4]

Deep- learning-assisted diagnosis for knee magnetic resonance imaging: development and retrospective validation of mrnet

Nicholas Bien, Pranav Rajpurkar, Robyn L Ball, Jeremy Irvin, Allison Park, Erik Jones, Michael Bereket, Bhavik N Patel, Kristen W Yeom, Katie Shpanskaya, et al. Deep- learning-assisted diagnosis for knee magnetic resonance imaging: development and retrospective validation of mrnet. PLoS medicine, 15(11):e1002699, 2018

work page 2018

[5] [5]

Stability and generalization

Olivier Bousquet and André Elisseeff. Stability and generalization. The Journal of Machine Learning Research, 2:499–526, 2002

work page 2002

[6] [6]

Grad-cam++: Generalized gradient-based visual explanations for deep convolu- tional networks

Aditya Chattopadhay, Anirban Sarkar, Prantik Howlader, and Vineeth N Balasubrama- nian. Grad-cam++: Generalized gradient-based visual explanations for deep convolu- tional networks. In 2018 IEEE winter conference on applications of computer vision (WACV), pages 839–847. IEEE, 2018

work page 2018

[7] [7]

How good is your ex- planation? algorithmic stability measures to assess the quality of explanations for deep neural networks

Thomas Fel, David Vigouroux, Rémi Cadène, and Thomas Serre. How good is your ex- planation? algorithmic stability measures to assess the quality of explanations for deep neural networks. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 720–730, 2022

work page 2022

[8] [8]

Efficient belief propagation for early vision

Pedro F Felzenszwalb and Daniel P Huttenlocher. Efficient belief propagation for early vision. International journal of computer vision, 70:41–54, 2006

work page 2006