Abstract: To efficiently visualize 3D irregular volumetric datasets,we present a novel approach for GPU-friendly regularization and rendering of tetrahedral volumetric datasets.The key idea is that a sparse tetrahedral volumetric data can be approximated by an octree with a limited depth,and the residual errors can be represented by a hashing table generated by the perfect spatial hashing technique.By converting the octree and the hashing table into an octree texture and a 3D lookup table respectively,a regular and array-wise representation for the underlying dataset is constructed.This dual-regularization reformulation not only allows for random access in GPU,but also makes the visualization of tetrahedral volumetric datasets be identical to rendering two 3D regular textures.Experimental results demonstrate that high quality can be achieved with much less memory consumption than previous approaches when handling sparse datasets.