⚡ 本页包含 AI 生成的分析内容，仅供参考
该论文提出了一种建模/渲染统一的3D高斯溅射处理器，通过局部感知的动态细粒度渲染引擎和局部优化的统一渲染工作流，显著减少了冗余计算和能量消耗。该处理器在统一可重构架构中实现了神经建模和高斯渲染，相比现有3D GS加速器，渲染吞吐量提升3.4倍，每帧能耗降低74.1%，建模延迟降低数个数量级。
*Equally Credited Authors (ECAs) Abstract A modeling/rendering unified 3D GS processor is proposed with: 1) A locality-aware dynamic fine-grained rendering engine for reduced redundant computation. 2) A locality-optimized unified rendering workflow to reduce EMA. 3) A unified reconfigurable architecture for neural modeling and Gaussian rendering with minimal area overhead. It achieves 3.4× higher rendering throughput and 74.1% lower energy per frame than SOTA 3D GS accelerators, and an orders-of-magnitude reduction in modeling latency. 3D Gaussian Splatting (3D GS) [1] is widely adopted in applications such as virtual reality due to its realistic modeling and efficient rendering. However, traditional 3D GS modeling relies on backpropagation-based optimization with up to 30,000 iterations, leading to slow modeling and limited generalization. To enable faster modeling, feedforward GS modeling
Hedi Wang*, Xiaoyu Feng*, Weichen Gao, Huazhong Yang, Yongpan Liu
Tsinghua University, Beijing, China