HumanDreamer-X: Photorealistic Single-image Human Avatars Reconstruction via Gaussian Restoration

arXiv — cs.CVThursday, November 13, 2025 at 5:00:00 AM
HumanDreamer-X is a groundbreaking framework that revolutionizes single-image human reconstruction, a task critical for digital human modeling. Traditional methods often struggle with geometric inconsistencies, leading to issues like fragmented limbs in 3D models. By integrating multi-view generation and reconstruction, HumanDreamer-X significantly enhances both geometric consistency and visual fidelity. The framework employs 3D Gaussian Splatting for initial geometry and appearance, while HumanFixer ensures photorealistic results. This innovation has resulted in a PSNR improvement of 16.45%, achieving a PSNR of 25.62 dB, underscoring its potential impact on various applications in AI and digital media.
— via World Pulse Now AI Editorial System

Was this article worth reading? Share it

Recommended Readings
RealisticDreamer: Guidance Score Distillation for Few-shot Gaussian Splatting
PositiveArtificial Intelligence
The paper titled 'RealisticDreamer: Guidance Score Distillation for Few-shot Gaussian Splatting' discusses a new framework called Guidance Score Distillation (GSD) aimed at improving 3D Gaussian Splatting (3DGS). This technique addresses the overfitting issue encountered with sparse training views by leveraging multi-view consistency from pretrained Video Diffusion Models (VDMs). The proposed method demonstrates superior performance across multiple datasets, enhancing the quality of real-time 3D scene rendering.
Motion Matters: Compact Gaussian Streaming for Free-Viewpoint Video Reconstruction
PositiveArtificial Intelligence
The article discusses the introduction of the Compact Gaussian Streaming (ComGS) framework for online free-viewpoint video reconstruction. This innovative approach addresses the limitations of existing methods that struggle with high storage requirements due to point-wise modeling. By utilizing keypoint-driven motion representation, ComGS models object-consistent Gaussian point motion, significantly reducing storage needs. The framework achieves over 159 times storage reduction compared to 3DGStream and 14 times compared to the QUEEN method, enhancing efficiency in video reconstruction.