{
  "paper_id": "14fb7eec10aff19c228b314a33dfb2aa",
  "paper_title": "Lightweight Explicit 3D Human Digitization via Normal Integration",
  "paper_authors": [
    "Jiaxuan Liu",
    "Jingyi Wu",
    "Ruiyang Jing",
    "Han Yu",
    "Jing Liu",
    "Liang Song"
  ],
  "publication_date": "2025-02-28",
  "publication_venue": "Sensors",
  "abstract": "In recent years, generating 3D human models from images has gained significant attention in 3D human reconstruction. However, deploying large neural network models in practical applications remains challenging, particularly on resource-constrained edge devices. This problem is primarily because large neural network models require significantly higher computational power, which imposes greater demands on hardware capabilities and inference time. To address this issue, we can optimize the network architecture to reduce the number of model parameters, thereby alleviating the heavy reliance on hardware resources. We propose a lightweight and efficient 3D human reconstruction model that balances reconstruction accuracy and computational cost. Specifically, our model integrates Dilated Convolutions and the Cross-Covariance Attention mechanism into its architecture to construct a lightweight generative network. This design effectively captures multi-scale information while significantly reducing model complexity. Additionally, we introduce an innovative loss function tailored to the geometric properties of normal maps. This loss function provides a more accurate measure of surface reconstruction quality and enhances the overall reconstruction performance. Experimental results show that, compared with existing methods, our approach reduces the number of training parameters by approximately 80% while maintaining the generated model's quality.",
  "keywords": [
    "three-dimensional human reconstruction",
    "normal map estimation",
    "a skinned multi-person linear model",
    "deep learning"
  ],
  "paper_institutions": [
    "Academy for Engineering and Technology, Fudan University",
    "Innovation Platform for Academicians of Hainan Province"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:46:59.830664Z",
  "template_tag": "simple",
  "summary": "### 摘要\n本文针对从单张图像重建高精度3D人体模型时，大型神经网络模型在资源受限边缘设备上部署困难的问题，提出了一种轻量级且高效的3D人体重建方法。该方法以SMPL-X参数化模型为先验，利用可微渲染生成前后视角的粗糙法线图，并输入到所设计的轻量级生成网络Lite-GN中，该网络结合了膨胀卷积与跨协方差注意力机制，在扩大感受野和捕获跨通道特征的同时显著降低模型复杂度。网络输出的精细法线图通过d-BiNI方法进行法线积分，并结合SMPL-X先验完成人体网格的补全与重建。在训练过程中，作者引入基于法线图几何特性的VGG感知损失与Huber空间损失共同约束法线生成，提升了表面细节的精度。实验采用THuman2.0作为训练集，在CAPE和Renderpeople数据集上采用Chamfer距离、P2S距离、法线差异等多种指标进行定量评估，并与PIFuHD、PaMIR、ICON、ECON等方法对比，结果表明该方法在保持竞争性重建质量的同时，将网络参数从约345.6M降至50M（约减少80%），模型权重从约1.35GB降至201.6MB，推理速度也显著提升。此外，本文还探索了面部和手部细节优化、文本驱动的纹理生成以及基于SCANimate的可动画化化身构建，验证了方法的扩展性。未来工作将针对极端姿态和复杂衣物的鲁棒性以及整体一致性进行改进。\n\n### 关键要点\n- 提出一种轻量级3D人体重建框架，利用SMPL-X先验和法线图积分生成高保真带衣物的人体模型。\n- 设计Lite-GN生成网络，采用膨胀卷积块和跨协方差注意力块，在扩大感受野的同时降低计算负担。\n- 引入VGG感知损失与Huber空间损失联合训练，共同约束法线图生成，提高表面几何精度。\n- 在CAPE数据集上，与PIFuHD、PaMIR、ICON和ECON相比，本方法的模型参数仅50M，权重文件201.6MB，推理时间12.8ms，优于对比方法。\n- 在CAPE数据集上的定量结果显示，本方法Chamfer距离为0.965，P2S为0.930，法线差异为0.0472，质量接近ECON-EX但显著更轻量。\n- 在Renderpeople数据集上，本方法在法线差异指标上达到0.0452，优于所有对比方法，体现了对复杂衣物的鲁棒性。\n- 消融实验证实，移除跨协方差注意力块、膨胀卷积块、VGG损失或Huber损失均导致重建精度下降，表明各组件均不可或缺。\n- 用户研究显示，43位参与者在10对模型对比中更偏好本方法的重建结果，验证了主观质量优势。\n- 人脸和手部区域替换为SMPL-X生成部件可提升细节，但会损失个性化特征，提示需要权衡。\n- 利用预训练扩散模型和文本提示可生成高质量人体表面纹理，丰富了重建结果的应用。\n- 结合SCANimate可将重建模型驱动为可动画化身，展示了在动画和虚拟现实中的适用性。\n- 当前方法在复杂姿态和衣物下仍可能出现肢体缺失或背面不一致的伪影，未来需提升鲁棒性。"
}