Gaussian Dynamic Convolution for Efficient Single-Image Segmentation

TL;DR

This paper introduces Gaussian dynamic convolution (GDC), a novel module inspired by human visual perception, that enhances single-image segmentation by efficiently aggregating contextual information, outperforming existing methods on multiple benchmarks.

Contribution

The paper proposes GDC, a new convolutional module that adaptively samples spatial regions based on Gaussian distribution, improving segmentation accuracy and feature richness.

Findings

GDC outperforms existing convolutions on Pascal-Context, Pascal-VOC 2012, and Cityscapes datasets.

GDC produces richer, more vivid features compared to other convolution methods.

GDC is versatile and can be integrated into various segmentation network architectures.

Abstract

Interactive single-image segmentation is ubiquitous in the scientific and commercial imaging software. In this work, we focus on the single-image segmentation problem only with some seeds such as scribbles. Inspired by the dynamic receptive field in the human being's visual system, we propose the Gaussian dynamic convolution (GDC) to fast and efficiently aggregate the contextual information for neural networks. The core idea is randomly selecting the spatial sampling area according to the Gaussian distribution offsets. Our GDC can be easily used as a module to build lightweight or complex segmentation networks. We adopt the proposed GDC to address the typical single-image segmentation tasks. Furthermore, we also build a Gaussian dynamic pyramid Pooling to show its potential and generality in common semantic segmentation. Experiments demonstrate that the GDC outperforms other existing convolutions on three benchmark segmentation datasets including Pascal-Context, Pascal-VOC 2012, and Cityscapes. Additional experiments are also conducted to illustrate that the GDC can produce richer and more vivid features compared with other convolutions. In general, our GDC is conducive to the convolutional neural networks to form an overall impression of the image.