MULTI-SCALE CONTEXT AGGREGATION BY DILATED CONVOLUTIONS

採用dilated convolution代替pooling層,關於dilated conviolution的原理,可見: http://blog.csdn.net/u011961856/article/details/77141761 the front end module: 輸入爲三通道padded彩色圖像,輸出爲輸出feature map爲​,output channels爲21.採用VG

>>阅读原文<<