We also leverage LION's latent spaces for shape interpolation and autoencoding. This enables multimodal voxel-guided synthesis and shapeĭenoising. Retraining the latent DDMs: We can efficiently fine-tune LION's encoders on voxelized or noisy inputs, Point cloud-based modeling, ideal for DDMs, with surface reconstruction, desired by artists.įlexibility: Since LION is set up as a VAE, it can be easily adapted for different tasks without ![]() Reduces synthesis noise and enables us to generate high-quality geometry. Fine-tuning SAP on data generated by LION's autoencoder Varying Output Types: Extending LION with Shape As Points (SAP) geometry reconstructionĪllows us to also output smooth meshes. Global shape latent variable in a hierarchical manner further boosts expressivity. Point cloud structure for our main latent representation. Because of that, we use latent points, this is, we keep a Principle, an ideal representation for DDMs. This is easier than training on potentiallyĬomplex point clouds directly, thereby improving expressivity. LION has multiple advantages:Įxpressivity: By mapping point clouds into regularized latent spaces, the DDMs in latent space areĮffectively tasked with learning a smoothed distribution. Synthesize smooth shapes as desired by artists. Importantly, we also demonstrate how to augment LION with modern surface reconstruction methods to Novel latent samples from the hierarchical latent DDMs and decoding back to the original pointĬloud space. ![]() The latent representations are predicted with point cloud processingĮncoders, and two latent DDMs are trained in these latent spaces. LION comprises a hierarchical latent space with a vector-valued global shape latent and another However, it is constructed as a VAE with DDMs in latent Similar to previous 3D DDMs in this setting, LION operates on point clouds. LION focuses on learning a 3D generative model directly from geometry data without image-based training. We introduce the Latent Point Diffusion Model (LION), a DDM for 3D shape generation. Shape As Points (SAP) is optionally used for mesh reconstruction. The latent points can be interpreted as a smoothed version of the input point cloud. Point-Voxel CNNs (PVCNN) with adaptive Group Normalization (Ada. LION is set up as a hierarchical point cloud VAE with denoising diffusion models over the shape latent and latent point distributions. We hope that LION provides a powerful tool for artists working withģD shapes due to its high-quality generation, flexibility, and surface reconstruction. Surface reconstruction techniques to generate smooth 3D meshes. We also demonstrate shape autoencoding and latent shape interpolation, and we augment LION with modern Use LION for different relevant tasks: LION excels at multimodal shape denoising and voxel-conditioned synthesis, and it can be adaptedįor text- and image-driven 3D generation. Furthermore, our VAE framework allows us to easily Experimentally, LIONĪchieves state-of-the-art generation performance on multiple ShapeNet benchmarks. ![]() ![]() On point clouds directly, while the point-structured latents are still ideally suited for DDM-based modeling. The hierarchical VAE approach boosts performance compared to DDMs that operate We train two hierarchical DDMs in these latent spaces. LION is set up as a variational autoencoder (VAE) withĪ hierarchical latent space that combines a global shape latent representation with a point-structured latent space. Hierarchical Latent Point Diffusion Model (LION) for 3D shape generation. Synthesis and shape interpolation, and (iii) the ability to output smooth surfaces or meshes. To advance 3D DDMs and make them usefulįor digital artists, we require (i) high generation quality, (ii) flexibility for manipulation and applications such as conditional Optically Stimulated Luminescence ( OSL dating) on examples in the Wadi Wisad region suggest they were built in two main pulses, one in the Late Neolithic about 8,500 years ago and one about 5,400 years ago during the Early Bronze Age-Chalcolithic.Denoising diffusion models (DDMs) have shown promising results in 3D point cloud synthesis. The kites and associated walls (called desert kites) are thought to be mass kill hunting tools wheels (circular stone arrangements with spokes) appear to be constructed for funerary or ritual use, and pendants are strings of burial cairns. They are classified into four main categories based on their shape: kites, meandering walls, wheels, and pendants. First brought to scholarly attention by RAF pilots flying over the desert shortly after the Arab revolt of 1916, the geoglyphs were made of stacks of basalt, between two to three slabs high. In the Black Desert of Jordan, ruins, inscriptions, and geoglyphs are called by the Bedouin tribes who live the Works of the Old Men. Hundreds of thousands of geoglyphs are known in or close to lava fields throughout the Arabian peninsula.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |