To improve segmentation performance, a novel neural network architecture(termed DFCN-DCRF) is proposed, which combines an RGB-D fully convolutionalneural network (DFCN) with a depth-sensitive fully-connected conditional randomfield (DCRF). First, a DFCN architecture which fuses depth information into theearly layers and applies dilated convolution for later contextual reasoning isdesigned. Then, a depth-sensitive fully-connected conditional random field(DCRF) is proposed and combined with the previous DFCN to refine thepreliminary result. Comparative experiments show that the proposed DFCN-DCRFhas the best performance compared with most state-of-the-art methods.
展开▼
