Rendering of dynamic 3d scenes is challenging because it is impossible to perform preprocessing to merge and simplify polygonal models, to precalculate visibility information. The dynamic behavior of objects (visibility change, movement) is causing command buffers rebuilding and rejecting of invisible objects often does not result in performance gains. We propose an adaptive method for visualizing dynamic scenes, which selects the most efficient method for recording and using command buffers and the number of hardware occlusion queries. Proposed adaptive method is based on the performance model, which performs an estimation of the execution time of the main stages of forward rendering. Testing results of the proposed method showed its effectiveness when rendering large dynamic scenes.

rendering, command buffer recording, occlusion culling, performance model of rendering

Gonakhchyan Vyacheslav Igorevich,  ,  ISP RAS