We present Lagrangian Hashing, a representation for neural fields combining the characteristics of fast training NeRF methods that rely on Eulerian grids (i.e.~InstantNGP), with those that employ points equipped with features as a way to represent information (e.g. 3D Gaussian Splatting or PointNeRF). We achieve this by incorporating a point-based representation into the high-resolution layers of the hierarchical hash tables of an InstantNGP representation. As our points are equipped with a field of influence, our representation can be interpreted as a mixture of Gaussians stored within the hash table. We propose a loss that encourages the movement of our Gaussians towards regions that require more representation budget to be sufficiently well represented. Our main finding is that our representation allows the reconstruction of signals using a more compact representation without compromising quality.
We present SinGAN-3D, a variation of the deep neural network architecture presented originally by SinGAN, for the generation of 3D contents, starting from a single three-dimensional voxelized model. Our network uses a pyramid of 3D convolutional networks to model the third dimension and exploits periodic activation functions to capture the latent structure of the input model. The approach can synthesize contents at different resolutions and aspect ratios, and can be extended to implement super resolution. To evaluate the performances of the proposed model we use the Single Image Freche’t distance, and the multiscale structural similarity index. The metrics highlight the similarities between the synthesised-three dimensional assets and their corresponding original template. An additional user study has also been conducted to assess the quality of the generated shapes. The code can be found at: https://github.com/zenos4mbu/SinGAN3D