Partial Convolution based Padding

• DOI: 10.48550/arxiv.1811.11718
• Publication Date: 2018-01-01
• Authors: Guilin Liu; Kevin J. Shih; Ting-Chun Wang; Fitsum A. Reda; Karan Sapra; Zhiding Yu; Andrew Tao; Bryan Catanzaro
• Keywords: Padding; Convolution (computer science)

In this paper, we present a simple yet effective padding scheme that can be used as drop-in module for existing convolutional neural networks. We call it partial convolution based padding, with the intuition padded region treated holes and original input non-holes. Specifically, during operation, results are re-weighted near image borders on ratios between area sliding window area. Extensive experiments various deep network models ImageNet classification semantic segmentation demonstrate proposed consistently outperforms standard zero better accuracy.