Panoramic Depth Estimation Algorithm Based on Attention Module

With the continuous progress and development of virtual reality technology, the panoramic image is more and more widely used. The refinement of the panoramic image requires clear and accurate depth map information. Traditional monocular panoramic depth estimation methods can not obtain accurate depth information due to the lack of available information on the original color image. However, the existing depth estimation methods based on deep learning are mostly for conventional planar images, and it is difficult to generate high-precision depth maps for detailed features of panoramic images. To solve this problem, we propose a panoramic depth estimation algorithm based on the attention module in this paper. This algorithm adds an attention module to the feature extraction convolution layer to extract salient features of panoramic images. Meanwhile, a joint loss function is designed to optimize the network according to the panoramic image depth estimation task. Experiments show that the method proposed in this paper effectively solves the untargeted detail extraction in the network training process and improves the image quality of the generated panoramic depth map.