3D Human Reconstruction and Generation

The field of 3D human reconstruction and generation is moving towards more efficient and accurate methods for reconstructing and generating high-quality 3D models from sparse and uncalibrated data. Recent developments have focused on feed-forward models that can reconstruct and interpolate 3D humans in real-time, as well as unified frameworks that encode geometry and appearance in a single latent space. These advancements have enabled the generation of high-fidelity 3D assets and have improved the quality of 3D stylization and synthesis. Notable papers in this area include Forge4D, which proposes a feed-forward 4D human reconstruction and interpolation model, and UniLat3D, which introduces a unified framework for single-stage 3D generation. Additionally, Stylos presents a single-forward 3D Gaussian framework for 3D style transfer that operates on unposed content. These papers demonstrate significant improvements in the field and have the potential to impact various downstream applications.

Sources

Forge4D: Feed-Forward 4D Human Reconstruction and Interpolation from Uncalibrated Sparse-view Videos

UniLat3D: Geometry-Appearance Unified Latents for Single-Stage 3D Generation

Stylos: Multi-View 3D Stylization with Single-Forward Gaussian Splatting

Built with on top of