Pith. sign in

REVIEW 4 major objections 6 minor 47 references

Identifying Bias in Deep Neural Networks Using Image Transforms

T0 review · 4 major / 6 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 The paper claims that applying wavelet transforms and median filtering to full images can reveal when a CNN's classification accuracy is driven by hidden background bias, eliminating the need to crop blank background regions for such a…

desk verdict A plausible but empirically thin extension of the authors' own blank-background bias work, worth a conditional read but not yet a reliable diagnostic. read the letter →

arxiv 2412.13079 v1 pith:SNY6ROZS submitted 2024-12-17 cs.CV cs.AIcs.LG

Sai Teja Erukude , Akhil Joshi , Lior Shamir This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords datasetbiasconvolutionalneuralnetworksimagetransformswavelettransformmedianfilterVGG16benchmarkdatasetsmedicalimaging
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

Deep neural networks can score high on benchmark image datasets by exploiting hidden background artifacts rather than the objects they are supposed to recognize. This paper claims to offer a simple screen for that kind of bias: apply a wavelet transform, a median filter, or both to the full images, retrain or test the same network, and compare accuracy. For natural datasets assembled from many independent sources, accuracy drops after the transforms. For controlled, single-source datasets previously shown to carry imaging-process bias, accuracy stays flat or rises. The stable-or-higher accuracy is the paper's indicator that classification is being driven by reproducible background signal, and the proposed reading is that the test can flag bias even when no blank background region is available to crop.

What carries the argument

The test is a comparison of classification accuracy across four versions of each dataset: the original images, images passed through a two-dimensional discrete wavelet transform (Haar or Daubechies), images smoothed by a 5×5 median filter, and images given the median filter followed by the wavelet transform. The wavelet–median combination is the carrier of the argument: median filtering removes salt-and-pepper noise while preserving edges, and the subsequent wavelet decomposition localizes the residual high-frequency structure, which the authors hold reveals the systematic signal left by the image-acquisition process while destroying the contextual content that supports genuine recognition. The decision rule is that if the transformed-image accuracy matches or exceeds the original accuracy, the CNN's predictions are likely being driven by background bias rather than by the objects of interest.

What would settle it

Take a natural multi-source dataset and add a per-class background artifact of controlled strength, such as a faint fixed watermark or a constant gradient in every image of each class; then apply the wavelet-plus-median test. If the claim is correct, the transformed-versus-original accuracy pattern should shift from 'drops' to 'stable or rises' as the artifact is strengthened; if the pattern tracks dataset type instead, the diagnostic is not actually measuring imaging-process bias.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that image transforms act asymmetrically on contextual object information and on the artifacts of the imaging process. Using the VGG16 architecture on six datasets, the authors show that Fourier transform reduces accuracy everywhere and is therefore unusable as a diagnostic. In contrast, Haar and Daubechies wavelet transforms, a 5×5 median filter, and the median-plus-wavelet combination lower accuracy on the natural datasets (Imagenette, Natural Images) while leaving accuracy unchanged or improved on the synthetic and hybrid datasets (CT Scans, COIL-20, Yale Faces, Caltech 20). The authors interpret the same experiments on 20×20 blank-background crops, where wavelet transforms raise the accuracy of crops from biased datasets, as direct evidence that the transform is exposing hidden background signal. The concluding claim is that accuracy on the transformed full images can distinguish contextual information from imaging-process bias and alert to bias without separating background from the original images.

Load-bearing premise

The method assumes that the different reaction of natural versus synthetic datasets to the transforms is caused by the presence or absence of imaging-process bias, and not by other systematic differences between the groups, such as color versus grayscale images, single-source versus multi-source collection, or the regularity of object poses.

Editorial extensions

If this is right

  • A researcher can screen a dataset for this class of bias without needing blank background crops: if accuracy after wavelet or median+wavelet transforms is comparable to or higher than original accuracy, the classifier is likely exploiting background signal.
  • The method is aimed at controlled-environment datasets such as medical images, where every pixel may carry diagnostic information and no clearly irrelevant background region exists to crop.
  • The test is a one-sided check: the paper states that a negative result does not guarantee the absence of bias, since other kinds of bias can still be present.
  • The Fourier transform is explicitly ruled out as a diagnostic, because it degrades accuracy across natural, synthetic, and hybrid datasets alike and cannot separate contextual from bias signal.

Reading between the lines

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

  • A decisive control the paper does not run would be to inject a tunable per-class artifact, such as a faint watermark or a fixed illumination gradient, into natural images and verify that the accuracy pattern moves with artifact strength; such an experiment would isolate imaging-process bias from dataset provenance.
  • The six datasets differ in more than bias: the natural sets are color, multi-source photographs, while the biased positives are grayscale or single-setup captures, so the claimed diagnostic could in principle be tracking low-level statistical differences instead of bias. Converting the synthetic datasets to color or graying the natural datasets would test this.
  • The proposed rule is a warning flag, not a debiasing procedure; if adopted, it would motivate a follow-up stage that removes or neutralizes the flagged background signal rather than merely reporting it.
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 method to detect dataset bias in CNNs without cropping blank background regions, by applying image transforms (Fourier transform, Haar/Daubechies wavelet transforms, median filtering, and their combinations) to full images. Experiments are performed with VGG16 on six datasets, categorized as natural (Imagenette, Natural Images), synthetic (CT Scans, COIL-20, Yale Faces), and hybrid (Caltech 20). The main observation is that wavelet and median-based transforms decrease classification accuracy on natural datasets but keep it stable or increase it on synthetic and hybrid datasets; Fourier transform decreases accuracy everywhere. From this, the paper concludes in Section 5 that stable or higher post-transform accuracy indicates that classification is driven by imaging-process background bias, and positions the method as a practical bias-alert tool when blank backgrounds are unavailable. The code is publicly available.

Significance. If the proposed rule were validated, it would offer a simple and inexpensive proxy for detecting hidden imaging-process bias in datasets where blank-background controls are impossible, which is a real need for medical imaging and other controlled-capture domains. The paper has concrete strengths: it uses established benchmark datasets, the transforms are clearly described, the decision rule is explicitly stated, and the code is released. However, the current evidence is preliminary: there are no error bars or repeated-seed experiments, the six datasets differ systematically in color, grayscale, object variability, and background structure, and the biased/unbiased labels are taken from prior blank-background studies that also motivated the rule. Because the central claim is an empirical decision rule, the missing statistical support and the absence of controls against confounds are load-bearing, not presentation issues.

major comments (4)
  1. [§5, Table 1] The decision rule in Section 5 rests on accuracy differences, but Table 1 reports only point estimates from a single run with no confidence intervals, error bars, or statistical tests. Several decisive comparisons are very close: CT Scans median+wavelet is 70.5% versus 70.0% original, and Caltech 20 median+wavelet is 36.0% versus 36.6% original, which is a decrease, contrary to the stated pattern. With a single 70/15/15 split and one training run, these differences are within plausible noise. The paper should report results over multiple random seeds and data splits, with mean, standard deviation, and paired significance tests (e.g., McNemar's test or a paired bootstrap) for each transform versus the original images.
  2. [§3.2, §5] The method is validated in a circular manner: the synthetic datasets are selected because they were already known from prior blank-background studies to contain bias, and the rule 'similar or higher accuracy after transforms implies bias' is then derived from the observed responses on these same datasets. No independent, out-of-sample validation is performed. The paper should either prospectively predict bias status for datasets not used to develop the rule, or at least apply the method to additional datasets whose bias status is not already documented, and report sensitivity and specificity under a pre-registered threshold.
  3. [§3.1–§3.3, §4] The observed differential response is confounded with dataset type: natural datasets are color images with heterogeneous backgrounds and high object variability, while the synthetic datasets are grayscale, single-source, and have controlled or uniform backgrounds. The conclusion that the response is caused by the presence or absence of imaging-process bias requires control experiments that separate this factor from low-level image statistics. Suggested controls include: applying the same transforms to grayscale versions of Imagenette and Natural Images, adding natural-like background variability to synthetic datasets, or artificially implanting a known acquisition artifact into a natural dataset and checking that the method flags it.
  4. [§3.4, §4] All experiments use a single architecture (VGG16) with one training protocol (40 epochs, Adam). The paper repeatedly speaks of 'CNNs' in general, but no evidence is given that the pattern holds across architectures, depths, or training settings. At minimum, the key comparisons in Table 1 should be repeated with one or two additional standard architectures (e.g., ResNet or a smaller CNN) and with changes in training budget, since the proposed method is meant to be a general diagnostic for CNN bias.
minor comments (6)
  1. [Figure 7] The caption says 'VGG15 architecture' but the text describes VGG16; the caption should be corrected.
  2. [§4.4] The sentence 'The table provides a higher resolution view of the results' appears twice in succession; remove the duplicate.
  3. [§4.3 and Figures 14, 16] The y-axis labels in Figures 14 and 16 say 'Wavelet Accuracy' where the bar labels are median-filtered and median+wavelet, respectively; the labels should match the transform being reported.
  4. [§3.1] The 'Natural Images' dataset from Roy et al. (2018) is used with the 'fruit' class discarded; a brief description of the class counts and image sizes would help reproducibility.
  5. [§4] There are several typos and grammatical errors throughout, including 'This transformations affect', 'the images are cannot be registered', and 'Wavelet transformation were generated'; a careful language edit is needed.
  6. [Table 1] Table 1 does not report the random-chance accuracy for each dataset, which is important for interpreting values like Caltech 20's 36.6% accuracy; adding a chance-level row would improve readability.

Circularity Check

1 steps flagged · score 6.0 of 10

Bias-detection rule is fitted to the same six datasets used for validation, so the claimed alert is post hoc rather than a prediction.

  1. fitted input called prediction [Section 3.2 (dataset selection) and Section 5 (decision rule); conclusion summarized in the Abstract]
    ""Therefore, these datasets can be used as datasets that are known as datasets that contain information leading to bias, and any method for bias identification needs to be sensitive to the bias in these datasets." [...] "If the classification accuracy with the transformed images is similar or higher than the classification accuracy when using the original images, it is an indication that the classification might be driven by bias.""

    The six datasets were selected because prior work (largely by the same authors) had already labeled them as biased (COIL-20, Yale Faces, CT Scans) or unbiased (Imagenette, Natural Images). The transforms were then screened: Fourier was rejected because it did not separate these labels, while wavelet/median were retained because their accuracy behavior did separate them. The Section 5 rule ('similar or higher accuracy implies bias') is the same separation stated as a test. Applying it to the datasets used to choose the transforms and threshold reproduces the input labels by construction; it is a post hoc description, not an independent prediction. No held-out dataset with unknown bias is tested, so the claimed alert is not supported as a generalizable prediction.

full rationale

The paper is an empirical proposal rather than a derivation; there are no equations that would make a result equal to its input by algebra. The main circularity concern is evaluative: the decision rule and the choice of transforms are selected post hoc from six datasets whose bias labels come from the same authors' earlier blank-background experiments. This makes the 'alert' on those datasets trivially consistent with the labels, and no independent dataset is used to show that the rule generalizes. However, the prior blank-background results are separate external experiments, not a construct of this paper, and the observed accuracy differences are genuine measurements; therefore the circularity is partial (the claimed validation reduces to the training set) rather than a definitional equivalence. The paper also includes a caveat that a negative test does not prove absence of bias, which qualifies but does not repair the missing out-of-sample test.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The central claim rests on an empirical association between dataset type (synthetic and controlled versus natural) and accuracy change under transforms. There are no fitted numeric parameters beyond a hand-chosen median filter window, but the decision rule is qualitative and post hoc. The key axioms are that synthetic datasets are biased, that natural datasets are not, that transforms separate contextual from artifact information, and that VGG16 represents CNNs generally.

free parameters (1)
  • median filter window size = 5x5
    Hand-chosen by the authors; the paper reports insensitivity to sizes 7x7 and 9x9, so it is not a critical free parameter.
assumptions (4)
  • ad hoc to paper The difference in accuracy between original and transformed images is attributable to dataset bias, not other image properties.
    Section 5 inverts observed patterns into a decision rule; no control experiments rule out confounding factors such as color, image statistics, or object variability.
  • ad hoc to paper The transforms affect contextual visual information differently than imaging-process artifacts.
    This is the central mechanism proposed in the abstract and Section 4, but it is not directly demonstrated; only accuracy changes are reported.
  • domain assumption Synthetic and controlled datasets used here are biased while natural datasets are not.
    Bias labels come from prior studies by the same group (Model and Shamir 2015; Dhar and Shamir 2021), not re-established in this paper.
  • domain assumption VGG16 is representative of CNNs in general.
    Only VGG16 is tested, yet conclusions are framed for CNNs broadly (Section 5).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Identifying Bias in Deep Neural Networks Using Image Transforms." pith.science (2026). https://pith.science/paper/SNY6ROZS

@misc{pith2026241213079,
  author       = {Pith},
  title        = {Pith review of: Identifying Bias in Deep Neural Networks Using Image Transforms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SNY6ROZS}},
  note         = {Machine review of arXiv:2412.13079}
}
read the original abstract

CNNs have become one of the most commonly used computational tool in the past two decades. One of the primary downsides of CNNs is that they work as a ``black box", where the user cannot necessarily know how the image data are analyzed, and therefore needs to rely on empirical evaluation to test the efficacy of a trained CNN. This can lead to hidden biases that affect the performance evaluation of neural networks, but are difficult to identify. Here we discuss examples of such hidden biases in common and widely used benchmark datasets, and propose techniques for identifying dataset biases that can affect the standard performance evaluation metrics. One effective approach to identify dataset bias is to perform image classification by using merely blank background parts of the original images. However, in some situations a blank background in the images is not available, making it more difficult to separate foreground or contextual information from the bias. To overcome this, we propose a method to identify dataset bias without the need to crop background information from the images. That method is based on applying several image transforms to the original images, including Fourier transform, wavelet transforms, median filter, and their combinations. These transforms were applied to recover background bias information that CNNs use to classify images. This transformations affect the contextual visual information in a different manner than it affects the systemic background bias. Therefore, the method can distinguish between contextual information and the bias, and alert on the presence of background bias even without the need to separate sub-images parts from the blank background of the original images. Code used in the experiments is publicly available.

Figures

Figures reproduced from arXiv: 2412.13079 by the authors.

Figure 1
Figure 1. Additionally, there are noticeable overlaps [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 1
Figure 1. X-rays images classified correctly by CNNs. a) Original X-ray, b) AlexNet, c) GoogleNet, d) [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. A 20×20 cropped background segment of a blank background part of the original image. That was done to all images in the dataset to create a new dataset of blank sub-images, as shown in Fig￾ure 3. When using just these seemingly blank parts of the images, the classification accuracy of numerous datasets was far higher than mere chance accuracy. For instance, [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figures from the paper (13 more)
Figure 3
Figure 3. Figure 3: Original images from Yale Faces B and the 20 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png]
Figure 5
Figure 5. Figure 5: Classification accuracy of CNN models trained and tested on seemingly blank sub-images taken from image background of several common im￾age benchmark datasets (Dhar and Shamir, 2021). et al., 2001; Belitskaya, 2018) contains 165 images of 15 subjects, with different fa…
Figure 6
Figure 6. Figure 6: Categorization of data. The data used in [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: VGG16 architecture. 4 Methods and results As demonstrated in Section 2, CNNs are capable of identifying the correct class even when they are trained and tested with seemingly blank background parts of the images. The presence of such bias can be identified by performin…
Figure 8
Figure 8. Figure 8: Classification accuracy when using the Fourier-transformed full images. 0 10 20 30 40 50 60 70 Imagenette Natural Images CT Scans COIL-20 Yale Faces Caltech 20 Accuracy (%) Cropped Accuracy Cropped Fourier Accuracy [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: Classification accuracy when using the Fourier-transformed small sub-images taken from the background. the Fourier-transformed images leaves it challenging to determine whether the drop in classification accu￾racy is due to the loss of relevant features or driven by th…
Figure 10
Figure 10. Figure 10: Example of an original image and the [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Example of an original image taken from ImageNette, and the Haar and Daubechies Discrete Wavelet transformed images. thetic, accuracy maintains a constant level of 100%, and its cropped image accuracy has increased from ∼27.8% to about ∼31.4%. 0 10 20 30 40 50 60 70 8…
Figure 12
Figure 12. Figure 12: Classification accuracy when using images [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]
Figure 13
Figure 13. Figure 13: Classification accuracy when using the Wavelet transformed cropped images In contrast, natural datasets made of images col￾lected from various sources typically do not contain bias in the imaging process, as images of the same class are rarely taken during the same se…
Figure 14
Figure 14. Figure 14: Classification accuracy when using the full [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]
Figure 15
Figure 15. Figure 15: Classification accuracy when using the sub-images of the blank background after applying median filter. ter. As before, the analysis was applied to the full im￾ages, as well as the cropped images. The accuracy for natural datasets, such as Imagenette and Natural im￾ag…
Figure 16
Figure 16. Figure 16: Classification accuracy of full images after [PITH_FULL_IMAGE:figures/full_fig_p012_16.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 40 canonical work pages

  1. [1]

    Agarwal, S., Singh, O., and Nagaria, D. (2017). Analysis and comparison of wavelet transforms for denoising mri image. Biomed Pharmacol J , 10(2). Article

    work page 2017

  2. [2]

    Alqaraawi, A., Schuessler, M., Wei , P., Costanza, E., and Berthouze, N. (2020). Evaluating saliency map explanations for convolutional neural networks: a user study. In Proceedings of the 25th International Conference on Intelligent User Interfaces , pages 275--285

    work page 2020

  3. [3]

    Arun, N., Gaw, N., Singh, P., Chang, K., Aggarwal, M., Chen, B., Hoebel, K., Gupta, S., Patel, J., Gidwani, M., et al. (2021). Assessing the trustworthiness of saliency maps for localizing abnormalities in medical imaging. Radiology: Artificial Intelligence , 3(6):e200267

    work page 2021

  4. [4]

    Babul, K. A. R., Sathish, R., and Pattanaik, M. (2024). Synthetic simplicity: Unveiling bias in medical data augmentation. In Proceedings of Data Engineering in Medical Imaging: Second MICCAI Workshop, DEMI 2024 , page 64. Springer Nature

    work page 2024

  5. [5]

    Ball, P. (2023). Is ai leading to a reproducibility crisis in science? Nature , 624:22--25

    work page 2023

  6. [6]

    Belitskaya, O. (2018). Yale face database - kaggle

    work page 2018

  7. [7]

    Buhrmester, V., Münch, D., and Arens, M. (2021). Analysis of explainers of black box deep neural networks for computer vision: A survey. Machine Learning and Knowledge Extraction , 3(4):966--989

    work page 2021

  8. [8]

    T., Cooley, J

    Cochran, W. T., Cooley, J. W., Favin, D. L., Helms, H. D., Kaenel, R. A., Lang, W. W., Maling, G. C., Nelson, D. E., Rader, C. M., and Welch, P. D. (1967). What is the fast fourier transform? Proceedings of the IEEE , 55(10):1664--1674

    work page 1967

Show all 47 references
  1. [9]

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. (2009). Imagenet: A large-scale hierarchical image database. In 2009 IEEE Conference on Computer Vision and Pattern Recognition , pages 248--255

    2009

  2. [10]

    and Parker, M

    Dhanani, S. and Parker, M. (2013). Introduction to digital filtering. In Dhanani, S. and Parker, M., editors, Digital Video Processing for Engineers , pages 19--28. Newnes, Oxford

    2013

  3. [11]

    and Shamir, L

    Dhar, S. and Shamir, L. (2021). Evaluation of the benchmark datasets for testing the efficacy of deep convolutional neural networks. Visual Informatics , 5

    2021

  4. [12]

    and Shamir, L

    Dhar, S. and Shamir, L. (2022). Systematic biases when using deep neural networks for annotating large catalogs of astronomical images. Astronomy and Computing , 38:100545

    2022

  5. [13]

    Fisher, R., Perkins, S., Walker, A., and Wolfart, E. (1996). Hypermedia image processing reference. England: John Wiley & Sons Ltd , pages 209--213

    1996

  6. [14]

    D., Jordache, A., Vasdani, M., and Deng, J

    Gavrilov, A. D., Jordache, A., Vasdani, M., and Deng, J. (2018). Preventing model overfitting and underfitting in convolutional neural networks. International Journal of Software Science and Computational Intelligence , 10(4):19--28

    2018

  7. [15]

    S., Belhumeur, P

    Georghiades, A. S., Belhumeur, P. N., and Kriegman, D. J. (2001). From few to many: Illumination cone models for face recognition under variable lighting and pose. IEEE transactions on pattern analysis and machine intelligence , 23(6):643--660

    2001

  8. [16]

    Griffin, G., Holub, A., Perona, P., et al. (2007). Caltech-256 object category dataset. Technical report, Technical Report 7694, California Institute of Technology Pasadena

    2007

  9. [17]

    Hashimoto, N., Fukushima, D., Koga, R., Takagi, Y., Ko, K., Kohno, K., Nakaguro, M., Nakamura, S., Hontani, H., and Takeuchi, I. (2020). Multi-scale domain-adversarial multiple-instance cnn for cancer subtype classification with unannotated histopathological images. In Proceed...

    2020

  10. [18]

    and Ntoutsi, E

    Iosifidis, V. and Ntoutsi, E. (2018). Dealing with bias via data augmentation in supervised learning scenarios. Jo Bates Paul D. Clough Robert J \"a schke , 24(11)

    2018

  11. [19]

    Jaipuria, N., Zhang, X., Bhasin, R., Arafa, M., Chakravarty, P., Shrivastava, S., Manglani, S., and Murali, V. N. (2020). Deflating dataset bias using synthetic data augmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops , page...

    2020

  12. [20]

    Kim, B., Seo, J., Jeon, S., Koo, J., Choe, J., and Jeon, T. (2019). Why are saliency maps noisy? cause of and solution to noisy saliency maps. In 2019 IEEE/CVF International Conference on Computer Vision Workshop , pages 4149--4157. IEEE

    2019

  13. [21]

    Lin, Y., Zhao, H., Ma, X., Tu, Y., and Wang, M. (2020). Adversarial attacks in modulation recognition with convolutional neural networks. IEEE Transactions on Reliability , 70(1):389--401

    2020

  14. [22]

    Majeed, T., Rashid, R., Ali, D., and Asaad, A. (2020). Issues associated with deploying cnn transfer learning to detect covid-19 from chest x-rays. Physical and Engineering Sciences in Medicine , 43:1289–--1303

    2020

  15. [23]

    M., and Miller, P

    McLaughlin, N., Del Rincon, J. M., and Miller, P. (2015). Data-augmentation for reducing dataset bias in person re-identification. In 12th IEEE International conference on advanced video and signal based surveillance , pages 1--6. IEEE

    2015

  16. [24]

    and Shamir, L

    Model, I. and Shamir, L. (2015). Comparison of data set bias in object recognition benchmarks. IEEE Access , 3:1953--1962

    2015

  17. [25]

    A., Nayar, S

    Nene, S. A., Nayar, S. K., Murase, H., et al. (1996). Columbia object image library (coil-20). Citeseer

    1996

  18. [26]

    Opencv image filtering documentation

    OpenCV (2024). Opencv image filtering documentation

    2024

  19. [27]

    and Zeebaree, D

    Othman, G. and Zeebaree, D. Q. (2020). The applications of discrete wavelet transform in image processing: A review. Journal of Soft Computing and Data Mining , 1(2):31--43

    2020

  20. [28]

    Pastaltzidis, I., Dimitriou, N., Quezada-Tavarez, K., Aidinlis, S., Marquenie, T., Gurzawska, A., and Tzovaras, D. (2022). Data augmentation for fairness-aware machine learning: Preventing algorithmic bias in law enforcement systems. In Proceedings of the 2022 ACM Conference o...

    2022

  21. [29]

    and Maragos, P

    Pertigkiozoglou, S. and Maragos, P. (2018). Detecting adversarial examples in convolutional neural networks. arXiv:1812.03303

    2018 arXiv

  22. [30]

    T., Wei, M

    Pfau, J., Young, A. T., Wei, M. L., and Keiser, M. J. (2019). Global saliency: aggregating saliency maps to assess dataset artefact bias. arXiv:1910.07604

    2019 arXiv

  23. [31]

    Pywavelets documentation

    PyWavelet Developers (2024). Pywavelets documentation

    2024

  24. [32]

    A., Sathish, R., Pattanaik, M., and Venkataramani, R

    Raja Babu, K. A., Sathish, R., Pattanaik, M., and Venkataramani, R. (2024). Synthetic simplicity: Unveiling bias in medical data augmentation. In MICCAI Workshop on Data Engineering in Medical Imaging , pages 64--72. Springer

    2024

  25. [33]

    L., Byttner, W., Krumnack, U., Wiedenroth, A., Schallner, L., and Shenk, J

    Richter, M. L., Byttner, W., Krumnack, U., Wiedenroth, A., Schallner, L., and Shenk, J. (2021). (input) size matters for cnn classifiers. In Proceedings of 30th International Conference on Artificial Neural Networks ICANN 2021: , Bratislava, Slovakia, September 14--17, 2021, ,...

    2021

  26. [34]

    Roy, P., Ghosh, S., Bhattacharya, S., and Pal, U. (2018). Effects of degradations on deep neural network architectures. arXiv preprint arXiv:1807.10108

    2018 arXiv

  27. [35]

    Santos, C. F. G. D. and Papa, J. P. (2022). Avoiding overfitting: A survey on regularization methods for convolutional neural networks. ACM Computing Surveys , 54(10s):1--25

    2022

  28. [36]

    Shamir, L. (2008). Evaluation of face datasets as tools for assessing the performance of face recognition methods. International Journal of Computer Vision , 79:225--230

    2008

  29. [37]

    M., Bouneffouf, D., Muthusamy, V., and Varshney, K

    Sharma, S., Zhang, Y., R \' os Aliaga, J. M., Bouneffouf, D., Muthusamy, V., and Varshney, K. R. (2020). Data augmentation for discrimination prevention and bias disambiguation. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society , pages 358--364

    2020

  30. [38]

    and Benitez, R

    Sikha, O. and Benitez, R. (2024). Deep feature discriminability as a diagnostic measure of overfitting in cnn models. In Science and Information Conference , pages 218--233. Springer

    2024

  31. [39]

    Simonyan, K. (2013). Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv:1312.6034

    2013 arXiv

  32. [40]

    and Zisserman, A

    Simonyan, K. and Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. ArXiv: 1409.1556

    2014 arXiv

  33. [41]

    Thanapol, P., Lavangnananda, K., Bouvry, P., Pinel, F., and Lepr \'e vost, F. (2020). Reducing overfitting and improving generalization in training convolutional neural network (cnn) under limited sample sizes in image recognition. In International Conference on Information Te...

    2020

  34. [42]

    Tommasi, T., Patricia, N., Caputo, B., and Tuytelaars, T. (2015). A deeper look at dataset bias. CoRR , abs/1505.01257

    2015 arXiv

  35. [43]

    and Efros, A

    Torralba, A. and Efros, A. A. (2011). Unbiased look at dataset bias. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition , pages 1521--1528. IEEE

    2011

  36. [44]

    K., and Zhu, A

    Uchida, S., Ide, S., Iwana, B. K., and Zhu, A. (2016). A further step to perfect accuracy by training cnn with larger data. In 2016 15th International Conference on Frontiers in Handwriting Recognition (ICFHR) , pages 405--410. IEEE

    2016

  37. [45]

    black box

    Wang, B., Ma, R., Kuang, J., and Zhang, Y. (2020). How decisions are made in brains: Unpack “black box” of cnn with ms. pac-man video game. IEEE Access , 8:142446--142458

    2020

  38. [46]

    Wang, J., Zhao, J., Yin, Q., Luo, X., Zheng, Y., Shi, Y.-Q., and Jha, S. K. (2021). Smsnet: A new deep convolutional neural network model for adversarial example detection. IEEE Transactions on Multimedia , 24:230--244

    2021

  39. [47]

    Winastwan, R. (2024). Class activation mapping (cam): Better interpretability in deep learning models

    2024

Pith tools

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