Multi-Scale Context Aggregation by Dilated Convolutions

TL;DR

This paper introduces a new convolutional module using dilated convolutions for dense prediction tasks like semantic segmentation, effectively capturing multi-scale context without resolution loss, and improves existing models' accuracy.

Contribution

The paper presents a novel dilated convolution-based module specifically designed for dense prediction, enhancing multi-scale context aggregation and model accuracy.

Findings

Dilated convolutions support exponential receptive field expansion without resolution loss.

The new module improves the accuracy of state-of-the-art semantic segmentation systems.

Simplifying adapted image classification networks can increase dense prediction accuracy.

Abstract

State-of-the-art models for semantic segmentation are based on adaptations of convolutional networks that had originally been designed for image classification. However, dense prediction and image classification are structurally different. In this work, we develop a new convolutional network module that is specifically designed for dense prediction. The presented module uses dilated convolutions to systematically aggregate multi-scale contextual information without losing resolution. The architecture is based on the fact that dilated convolutions support exponential expansion of the receptive field without loss of resolution or coverage. We show that the presented context module increases the accuracy of state-of-the-art semantic segmentation systems. In addition, we examine the adaptation of image classification networks to dense prediction and show that simplifying the adapted network can increase accuracy.