Significance Recently, super multi-view light-field display devices have made great progress, showing the characteristics of increasing views, higher resolution, and larger viewing angles. The virtual scene to be displayed is becoming increasingly complex.

Herein, we review and summarize the existing virtual stereo content generation technology for super multiview light-field display devices. Their applicable scenarios, merits, and defects are illustrated. The main challenges of content generation of exisistng super multiview light-field display devices are highlighted. Real-time light-field content generation tehcnology for large-scale virtual scenes of super multiview is not fully mature. A new rendering system needs to be explored to break through the bottleneck of existing algorithms. With the technical breakthrough, the popularization and application of light-field in key fields such as military affairs, agriculture, and fire protection could be solved.

Progress Recently, the super multiview virtual content generation technology has made consistent progress in the following three aspects, which are summarized in table 2.

Graphic processing unit (GPU) instancing is used to improve the view-by-view light-field rendering technology (Fig. 18). When there are many repetitive objects in the virtual scene that needs to be drawn, the GPU instancing technology only requires to set the original data buffer once and then draw calls. Compared with the original view-by-view rendering algorithm based on geometric shaders, as the number of views increases, GPU instancing requires no more memory. When there is no object repeatedly drawn in the virtual scene, the super multiview light-field image of a large-scale scene can be rendered simultaneously.

The delay shading technique is introduced to improve the super multiview light-field rendering technique based on depth rendering (Fig. 23). Depth-based calculations can produce false illumination information, which is considerably different from the illumination characteristic in real space. The three-dimensional technology team of the Beijing University of Posts and Telecommunications proposed an image generation method of super multiview light-field based on delay coloring in 2020. Its reference image comprises one or more pairs of color, depth, highlight, and environment images. Although the lighting problem is solved using this method, it still leads to the degradation of multiview image quality and depth-based rendering.

An ultrafast experimental light-field rendering pipeline is constructed using geometric correlation (Fig. 24). In 2020, a super multiview rendering pipeline for polygon rasterization based on the principle of geometric correlation was developed. In the case of a 2×10⁸ surface slice with a total resolution of 7680×4320, the rendering rate reached 60 frame/s. Other factors have not been considered, such as lighting model and texture mapping, so it cannot be compatible with the traditional hardware rendering pipeline.

Conclusions and Prospect Many rendering algorithms can be used to generate super multiview light-field content. The ray tracing algorithm has the highest rendering quality, which can be used for nonreal-time applications. The algorithm with the fastest speed is the rendering algorithm based on geometric correlation; however, it is incompatible with the existing hardware rendering pipeline. Therefore, that lighting and occlusion need to be treated specially in the rendering process. Currently, the two most practical real-time rendering technologies are the rendering algorithm based on GPU instancing and the deferred rendering method. The former could degrade the image quality, whereas the latter is not fully competent for high-quality rendering of large-scale scenes. In general, the real-time light-field content generation technology for large-scale virtual scenes of super multiview is not yet fully developed. Therefore, it is still necessary to explore a new rendering system to break through the bottlenecks of existing algorithms.