Image generation using one or more neural networks

摘要

One or more neural networks are used to generate one or more three-dimensional (3D) images 556 based, at least in part, on two or more two-dimensional (2D) images having different frames of reference or viewpoints 552. The neural networks may include feature extraction networks to extract semantic features from the 2D images (Fig. 5A, 504). Frustum-shaped clouds may be generated for the 2D images (Fig. 5A, 506) and may then further be placed within the 3D world frame of reference (Fig. 5A, 508). Trajectories moving through the captured environment may be determined and further represented within the 3D image.