Abstract
Producing 3D human representations from input views on the fly is essential for immersive live streaming systems, where representation compactness is as critical as high fidelity due to limited computational power and transmission bandwidth. Although recent feed-forward reconstruction methods achieve impressive quality through the view-centric prediction of 3D representations, they repeatedly encode the same subject content across multiple views, leading to significant inter-view redundancy. Our key insight is to perform predictions directly within 3D space, enabling the network to learn and produce a highly compact representation. To this end, we propose PointSplat, a novel human-centric approach that directly infers Gaussian primitives from an input point set. The proposed method first estimates a coarse geometric proxy and performs ray-casting to prune redundant points and establish explicit 2D–3D correspondences. Subsequently, it employs a Point-Image Transformer to fuse appearance and geometry features, predicting Gaussian attributes in a single forward pass. This design restricts predictions to foreground regions of interest, substantially reducing the total number of Gaussians while improving the rendering quality of novel views. Extensive experiments demonstrate that PointSplat achieves higher efficiency and quality, while exhibiting strong robustness to variations in view count and image resolution across multiple datasets.
