W

Wavelet Transform

Plenoxels-based method for constructing a radiance field for view synthesis by optimizing the coefficients of a three-dimensional grid representing the radiance field in a wavelet basis. Previous approaches based on spatial discretization have demonstrated the ability to efficiently optimize radiance fields without the use of a multilayer perception, significantly reducing optimization time compared to neural network-based methods. From other methods of three-dimensional reconstruction based on radiance fields that use frequency analysis tools, the presented method differs not only in not requiring a multilayer perceptron but also in not requiring a reduction of dimensions (i.e. from 3D to 2D) before performing the wavelet transform.

CUDA-accelerated implementation of Partial 3D Discrete Wavelet Transform, enabling efficient multi-resolution analysis of volumetric data. Designed optimized GPU kernels to support partial inverse transform, allowing reconstruction of arbitrary points in a 3D signal without processing the entire dataset.