mlsanityData to 7 Oct 2026

Mind the Pad -- CNNs can Develop Blind Spots

arXiv 2010.02178 ↗First posted 5 Oct 2020cs.CV cs.AI stat.MLcited by 41

Not on the map. Why?

Abstract (arXiv)

We show how feature maps in convolutional networks are susceptible to spatial bias. Due to a combination of architectural choices, the activation at certain locations is systematically elevated or weakened. The major source of this bias is the padding mechanism. Depending on several aspects of convolution arithmetic, this mechanism can apply the padding unevenly, leading to asymmetries in the learned weights. We demonstrate how such bias can be detrimental to certain tasks such as small object detection: the activation is suppressed if the stimulus lies in the impacted area, leading to blind spots and misdetection. We propose solutions to mitigate spatial bias and demonstrate how they can improve model accuracy.

Key facts. Cited by 41 papers in the map (data through 7 Oct 2026). First posted on arXiv 5 Oct 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 12; 2022: 15; 2023: 8; 2024: 1; 2025: 3; 2026: 2.

41 papers.

Sort:

By area:

  1. 2022Simple Baselines for Image Restorationcited by 701
  2. 2022Long Video Generation with Time-Agnostic VQGAN and Time-Sensitive Transformercited by 244
  3. 2022Feature-Realistic Neural Fusion for Real-Time, Open Set Scene Understandingcited by 37
  4. 2022Stabilizing Off-Policy Deep Reinforcement Learning from Pixelscited by 32
  5. 2023Deep Image Fingerprint: Towards Low Budget Synthetic Image Detection and Model Lineage Analysiscited by 29
  6. 2021Position, Padding and Predictions: A Deeper Look at Position Information in CNNscited by 26
  7. 2021FoveaTer: Foveated Transformer for Image Classificationcited by 26
  8. 2022High-fidelity GAN Inversion with Padding Spacecited by 23
  9. 2021GANSeg: Learning to Segment by Unsupervised Hierarchical Image Generationcited by 19
  10. 2021Toward Spatially Unbiased Generative Modelscited by 17
  11. 2021Shift Invariance Can Reduce Adversarial Robustnesscited by 16
  12. 2021Effects of boundary conditions in fully convolutional networks for learning spatio-temporal dynamicscited by 14
29 more
  1. 2022Classification of Long Sequential Data using Circular Dilated Convolutional Neural Networkscited by 13
  2. 2021Global Pooling, More than Meets the Eye: Position Information is Encoded Channel-Wise in CNNscited by 12
  3. 2022Cylin-Painting: Seamless 360textdegree Panoramic Image Outpainting and Beyondcited by 12
  4. 2022Finding Biological Plausibility for Adversarially Robust Features via Metameric Taskscited by 11
  5. 2023RecRecNet: Rectangling Rectified Wide-Angle Images by Thin-Plate Spline Model and DoF-based Curriculum Learningcited by 8
  6. 2025Foundation Models and Transformers for Anomaly Detection: A Surveycited by 8
  7. 2021How Shift Equivariance Impacts Metric Learning for Instance Segmentationcited by 7
  8. 2022Prescriptive and Descriptive Approaches to Machine-Learning Transparencycited by 7
  9. 2023Zone Evaluation: Revealing Spatial Bias in Object Detectioncited by 7
  10. 2023Hierarchical Disentanglement-Alignment Network for Robust SAR Vehicle Recognitioncited by 6
  11. 2021NeRD: Neural Representation of Distribution for Medical Image Segmentationcited by 5
  12. 2021LatentKeypointGAN: Controlling Images via Latent Keypointscited by 5
  13. 2022Unveiling The Mask of Position-Information Pattern Through the Mist of Image Featurescited by 3
  14. 2022Generalization to translation shifts: a study in architectures and augmentationscited by 3
  15. 2021Improved Padding in CNNs for Quantitative Susceptibility Mappingcited by 2
  16. 2021Seamless Satellite-image Synthesiscited by 2
  17. 2026Auditing Sybil: Explaining Deep Lung Cancer Risk Prediction Through Generative Interventional Attributionscited by 2
  18. 2026AdaSpot: Spend Resolution Where It Matters for Precise Event Spottingcited by 2
  19. 2022Super-Resolution Based Patch-Free 3D Image Segmentation with High-Frequency Guidancecited by 1
  20. 2023On the Interplay of Convolutional Padding and Adversarial Robustnesscited by 1
  21. 2023Padding-free Convolution based on Preservation of Differential Characteristics of Kernelscited by 1
  22. 2025On Symmetries in Convolutional Weightscited by 1
  23. 2022Exoplanet Cartography using Convolutional Neural Networkscited by 0
  24. 2022LatentKeypointGAN: Controlling Images via Latent Keypoints -- Extended Abstractcited by 0
  25. 2022Shift Variance in Scene Text Detectioncited by 0
  26. 2023Improving Translation Invariance in Convolutional Neural Networks with Peripheral Prediction Paddingcited by 0
  27. 2023Padding Aware Neuronscited by 0
  28. 2024Certified ML Object Detection for Surveillance Missionscited by 0
  29. 2025A Systematic Reproducibility Study of BSARec for Sequential Recommendationcited by 0