HYBRID NETWORK ARCHITECTURE FOR CALCULATING THE MONOCULAR IMAGE DEPTH
The article suggests a new architecture of a deep convoluted network for calculating the monocular image depth that combines convolution and recurrent layers. In the proposed architecture, convolution layers are used to calculate high-level feature maps, and recurrent layers, capturing global context information, improve the quality of feature maps. Experimental research showed that the proposed architecture demonstrates better results than the up-to-date techniques and does not require image post-processing.
Keywords: neural network, image processing, convolution layers, recurrent network
