Mind the Pad -- CNNs Can Develop Blind Spots

Authors: Bilal Alsallakh, Narine Kokhlikyan, Vivek Miglani, Jun Yuan, Orion Reblitz-Richardson

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 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.
Researcher Affiliation Collaboration Bilal Alsallakh Facebook AI Narine Kokhlikyan Facebook AI Vivek Miglani Facebook AI Jun Yuan NYU Orion Reblitz-Richardson Facebook AI
Pseudocode No The paper mentions "code in supplemental" for certain computations but does not present any pseudocode or algorithm blocks.
Open Source Code Yes The scripts used to generate the visualizations in this paper are available in the supplemental as well as at http://mind-the-pad.github.io.
Open Datasets Yes The model is trained on the BSTLD dataset [4] which annotates traffic lights in road scenes. ... Image Net classifiers based on Res Net [12] and Mobile Net [13].
Dataset Splits No The paper mentions training and testing but does not explicitly specify validation splits or proportions for training, validation, and test sets.
Hardware Specification No The paper describes experiments but does not provide specific details about the hardware used, such as GPU models or CPU specifications.
Software Dependencies No The paper mentions using TensorFlow [1] and PyTorch [33] but does not specify their version numbers or other software dependencies with versions.
Experiment Setup No The paper describes various experimental conditions like different input sizes (224x224, 225x225) and padding methods, but it lacks specific hyperparameters (e.g., learning rate, batch size, optimizer) needed for reproduction.