SMVRT：基于稀疏多视角体素重建与 Transformer 融合的隐式人体三维建模
SMVRT: Implicit Human 3D Modeling Using Sparse Multi-View Volumetric Reconstruction with Transformer Fusion
在稀疏多视角人体重建中，用三阶段注意力融合（2D 特征增强、warp 后多视角多层级特征融合成 3D 特征网格、2D-3D 查询点聚合）替代生成式补全或模板拟合来推断遮挡区域，这套融合模块可直接迁移到稀疏视角的通用隐式重建与占据预测流程。
原文依据 attentional fusion is performed on warped multi-view and multi-level 2D features to form 3D feature grid
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近年来，从单视角或多视角输入进行人体建模取得了显著进展。然而，这些方法往往依赖生成模型或模板拟合来猜测被遮挡区域。本文通过探索仅使用稀疏多视角输入的最优融合策略来应对这些挑战。我们提出 SMVRT，一个面向稀疏多视角人体建模的端到端隐式三维重建框架。我们的核心贡献在于网络三个阶段的融合模块。首先，设计局部与全局特征交替融合模块以增强二维特征。其次，对经 warp 的多视角、多层级二维特征执行注意力融合，形成三维特征网格；该特征网格通过三维正则化聚合空间一致的多视角特征。第三，注意力式二维-三维特征聚合为查询点生成增强的潜变量嵌入，以解码占据情况。在 THUman 2.0/2.1、MultiGarment 和 MultiHuman 数据集上的实验表明，我们的系统在定性和定量上均显著优于最优方法。
英文原文摘要 1154 字符
Recently, the community has witnessed significant progress in human modeling from single or multi-view inputs. However, these approaches often rely on guessing the occluded regions through either generative models or template fitting. In this work, we address these challenges by exploring optimal fusion strategies using only sparse multi-view inputs. We propose SMVRT, an end-to-end implicit 3D reconstruction framework for sparse multi-view human modeling. Our key contribution lies in the fusion blocks at three stages of the network. First, local and global features alternative fusion modules are designed to enhance 2D features. Second, attentional fusion is performed on warped multi-view and multi-level 2D features to form 3D feature grid. The feature grid aggregates spatially coherent multi-view features by 3D regularization. Third, attentional 2D-3D feature aggregation generates the enhanced latent embeddings for query points to decode occupancies. Experiments on the THUman 2.0/2.1, MultiGarment and MultiHuman datasets demonstrate that our system significantly outperforms state-of-the-art methods both qualitatively and quantitatively.