Monocular Depth Estimation Network Based on Swin Transformer

Abstract Estimating depth from a single image is challenging because 2D may correspond to many different 3D scenes with the same depth. While most deep learning based prediction methods extract features using small convolutional kernels receptive fields, which results in deformed edges and inaccurate values of distant objects estimation results. Aiming at this problem, we propose network on Swin Transformer encoder-decoder structure. We construct encoder network, can encode long-range spatial dependency various scales across channels. The decoder proposed charge fusing by operations interpolation, concatenation, convolution. Experiments KITTI NYUv2 datasets show that our get more accurate than state-of-the-art methods.