Triangle Grid Neural Radiation Field Model Compression for Space Scene Modeling
To address the issue of excessive memory usage in neural radiance field rendering meth-ods,a neural triangular mesh algorithm has been proposed.This algorithm constructs vertex feature vectors within trained 3D mesh models and utilizes rasterization in the traditional rendering pipeline to calculate the intersection of rays with the triangular mesh.It obtains the characteristicsof the cor-responding intersection points through barycentric coordinates.Experimental results on multiple pub-lic datasets demonstrate that this method requires only a small amount of storage space to store the neural radiance field model while ensuring rendering quality,achieving a model compression rate of 25%.Visualized experimental results indicate that this method can produce high-quality novel view synthesis results.The approach can eliminate the features of useless points in the scene and within the scene itself,reducing the model,s storage space.Moreover,the vertex features of the triangular mesh can be directly accessed using vertex indices,making it applicable on embedded devices.
model compressionneural radiation fieldraster renderingneural triangle grid
万方数据
维普
