实时语义分割网络结构汇总

为激发语义分割研究的新想法，特将最近的实时语义分割网络结构进行了汇总，方便大家对实时语义分割网络结构有一个直观快速的了解。

Design of Real-time Semantic Segmentation Decoder for Automated Driving

An efficient solution for semantic segmentation: ShuffleNet V2 with atrous separable convolutions

BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation

CGNet: A Light-weight Context Guided Network for Semantic Segmentation

Concentrated-Comprehensive Convolutions for lightweight semantic segmentation

ContextNet: Exploring Context and Detail for Semantic Segmentation in Real-time

DFANet: Deep Feature Aggregation for Real-Time Semantic Segmentation

DSNet for Real-Time Driving Scene Semantic Segmentation

Efficient ConvNet for Real-time Semantic Segmentation

Efficient Dense Modules of Asymmetric Convolution for Real-Time Semantic Segmentation

ERFNet: Efficient Residual Factorized ConvNet for Real-time Semantic Segmentation

ESPNet: Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation

ShelfNet for Real-time Semantic Segmentation

ESPNetv2: A Light-weight, Power Efficient, and General Purpose Convolutional Neural Network

Fast-SCNN: Fast Semantic Segmentation Network

反思

仔细观察以上各个实时语义分割网络的结构，发现其实万变不离其宗，都可以把他们归结到我的这篇博客[https://blog.csdn.net/sinat_24674017/article/details/96587343]总结的技巧。同时各个实时语义分割算法在benchmark上测试的结果也多是速度和mIoU的trade-off。从以上总结也可以看出目前网络结构的设计上已经没有太多创新之处，多是各种变异卷积和跳跃连接的堆积，这也让人产生怀疑：目前的深度学习研究是否多为新瓶装旧酒，已经遇到了瓶颈，难以再有更大的突破？