EVENT DOSSIER
The researchers proposed a new method based on Temporal Residual Neural Radiance Fields, aimed at solving the problem of 3D human modeling in dynamic scenarios. This method optimizes the temporal signal representation in video sequences by constructing temporal residual fields that are independent of the MLP architecture. Additionally, an integrated approach reduces the number of trainable parameters, thereby accelerating the rendering process and enhancing feature representation. Moreover, the designed multi-dimensional loss function accurately measures the difference between the predicted results and the actual spatial pixel values. Experimental results show that this method outperforms the latest representative methods in terms of PSNR and SSIM metrics. While maintaining similar accuracy to Anim-NeRF and Neural Body, its time efficiency is increased by nearly 780 times.
arXiv:2609.04984v1 Temporal Residual Neural Radiance Fields 提出动态人体重建新方法
本文提出一种基于时序残差神经辐射场（Temporal Residual Neural Radiance Fields）的方法，用于动态场景下的人体三维建模。该方法通过构建与时序 MLP 架构无关的时序残差场解决视频序列中的时序信号表示问题；同时采用集成方案减少可训练参数以加速渲染并增强特征表达能力；最后设计多维损失函数以准确衡量预测与真实空间像素值的差异。实验结果表明，该方法在峰值信噪比（PSNR）和结构相似性指数（SSIM）指标上优于最新代表性方法，在与 Anim-NeRF 和 Neural Body 保持相似精度的同时，时间效率提升了近 780 倍。