The Beautiful Future
tensorflow deeplab semantic segmentation 분석 본문
싸이트
https://github.com/tensorflow/models/tree/master/research/deeplab
레퍼런스
[1] Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs, ICLR2015
[2] DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs, TPAMI2017
[3] Rethinking Atrous Convolution for Semantic Image Segmentation, arXiv
[4] Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation, arXiv
1. 사용된 기법
- atrous convolution을 사용함. [1]
- atrous convolution pyramid pooling(ASPP)을 사용함. [2]
- ASPP에 image-level feature를 augment함, batch normalization을 사용,
학습과 평가에서 다른 출력 스트라이드를 사용했다. 학습: 16, 평가: 8. [3]
Comments