mlsanityData to 7 Oct 2026

Fix your downsampling ASAP! Be natively more robust via Aliasing and Spectral Artifact free Pooling

arXiv 2307.09804 ↗First posted 19 Jul 2023cs.CV eess.IVcited by 0

Not on the map. Why?

Abstract (arXiv)

Convolutional Neural Networks (CNNs) are successful in various computer vision tasks. From an image and signal processing point of view, this success is counter-intuitive, as the inherent spatial pyramid design of most CNNs is apparently violating basic signal processing laws, i.e. the Sampling Theorem in their downsampling operations. This issue has been broadly neglected until recent work in the context of adversarial attacks and distribution shifts showed that there is a strong correlation between the vulnerability of CNNs and aliasing artifacts induced by bandlimit-violating downsampling. As a remedy, we propose an alias-free downsampling operation in the frequency domain, denoted Frequency Low Cut Pooling (FLC Pooling) which we further extend to Aliasing and Sinc Artifact-free Pooling (ASAP). ASAP is alias-free and removes further artifacts from sinc-interpolation. Our experimental evaluation on ImageNet-1k, ImageNet-C and CIFAR datasets on various CNN architectures demonstrates that networks using FLC Pooling and ASAP as downsampling methods learn more stable features as measured by their robustness against common corruptions and adversarial attacks, while maintaining a clean accuracy similar to the respective baseline models.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 38 of its 58 references that are in the map. First posted on arXiv 19 Jul 2023.

Builds on

38 of 58 references.

References · 38

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  3. 2014Explaining and Harnessing Adversarial Examplescited by 7,287
  4. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,326
  5. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
  6. 2016Wide Residual Networkscited by 3,057
  7. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,857
  8. 2019Theoretically Principled Trade-off between Robustness and Accuracycited by 1,510
  9. 2020Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attackscited by 1,442
  10. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,430
  11. 2021Alias-Free Generative Adversarial Networkscited by 1,181
  12. 2016Adversarial Machine Learning at Scalecited by 934
26 more
  1. 2019AugMix: A Simple Data Processing Method to Improve Robustness and Uncertaintycited by 897
  2. 2018AutoAugment: Learning Augmentation Policies from Datacited by 728
  3. 2020Fast is better than free: Revisiting adversarial trainingcited by 723
  4. 2019Square Attack: a query-efficient black-box adversarial attack via random searchcited by 682
  5. 2020RobustBench: a standardized adversarial robustness benchmarkcited by 659
  6. 2019Unlabeled Data Improves Adversarial Robustnesscited by 415
  7. 2019Making Convolutional Networks Shift-Invariant Againcited by 390
  8. 2020Do Adversarially Robust ImageNet Models Transfer Better?cited by 315
  9. 2020Watch your Up-Convolution: CNN Based Generative Deep Neural Networks are Failing to Reproduce Spectral Distributionscited by 278
  10. 2020Uncovering the Limits of Adversarial Training against Norm-Bounded Adversarial Examplescited by 275
  11. 2021Improving Robustness using Generated Datacited by 233
  12. 2021Fixing Data Augmentation to Improve Adversarial Robustnesscited by 201
  13. 2018Decoupling Direction and Norm for Efficient Gradient-Based L2 Adversarial Attacks and Defensescited by 109
  14. 2020Understanding Catastrophic Overfitting in Single-step Adversarial Trainingcited by 75
  15. 2021On the effectiveness of adversarial training against common corruptionscited by 71
  16. 2020Wavelet Integrated CNNs for Noise-Robust Image Classificationcited by 60
  17. 2020Delving Deeper into Anti-aliasing in ConvNetscited by 60
  18. 2021Mind the box: l1-APGD for sparse adversarial attacks on image classifierscited by 50
  19. 2020Single-step Adversarial training with Dropout Schedulingcited by 40
  20. 2020Spectral Distribution Aware Image Generationcited by 29
  21. 2021WaveCNet: Wavelet Integrated CNNs to Suppress Aliasing Effect for Noise-Robust Image Classificationcited by 29
  22. 2022FrequencyLowCut Pooling -- Plug & Play against Catastrophic Overfittingcited by 28
  23. 2021Impact of Aliasing on Generalization in Deep Convolutional Networkscited by 25
  24. 2021Spectral Leakage and Rethinking the Kernel Size in CNNscited by 11
  25. 2021Anti-aliasing Deep Image Classifiers using Novel Depth Adaptive Blurring and Activation Functioncited by 11
  26. 2022Aliasing is a Driver of Adversarial Attackscited by 1

Built on by

Not cited yet.