Mind the Pad -- CNNs can Develop Blind Spots
Bilal Alsallakh, Narine Kokhlikyan, Vivek Miglani, Jun Yuan, and Orion Reblitz-Richardson

TL;DR
This paper reveals that padding in CNNs introduces spatial bias, causing blind spots that impair small object detection, and proposes solutions to mitigate this bias and improve accuracy.
Contribution
It identifies padding as a key source of spatial bias in CNNs and offers mitigation strategies to enhance model performance on biased tasks.
Findings
Padding causes systematic spatial bias in CNN feature maps.
Bias leads to blind spots and reduces detection accuracy for small objects.
Proposed solutions effectively mitigate bias and improve model performance.
Abstract
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.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
Videos
Taxonomy
TopicsAdvanced Neural Network Applications · Adversarial Robustness in Machine Learning · Domain Adaptation and Few-Shot Learning
MethodsConvolution
