Multi-Scale Context Aggregation by Dilated Convolutions
Fisher Yu Princeton University \ANDVladlen Koltun Intel Labs
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 problems such as semantic segmentation are structurally different from image classification. 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.
中文速览
语义分割（semantic segmentation）长期借用图像分类网络，但分类网络靠反复下采样压缩分辨率来获取全局感受野，直接拿来做逐像素预测会损失大量空间细节。本文提出用膨胀卷积（dilated convolution）搭建一个专为密集预测设计的上下文聚合模块（context module）：通过指数级递增的膨胀系数，感受野可以成倍扩大，同时始终保持全分辨率输出，无需任何下采样或多尺度输入。作者还重新审视了将分类网络迁移到分割任务的惯常做法，去掉了拖累性能的多余池化层，得到一个比FCN和DeepLab都更简洁、准确率高出5个百分点以上的基础预测模块；在此基础上接入上下文模块后，在Pascal VOC 2012测试集上进一步提升到75.3% mean IoU，超越当时所有方法。这项工作表明，专门为密集预测量身设计的网络结构，比从分类网络"改造"而来的方案更具优势，膨胀卷积此后也成为语义分割领域的基础构件。
原文 arXiv:1511.07122;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1511.07122v3