2025-02-14 01:50:52 +08:00
|
|
|
|
# Gaussian Haircut:使用股线对齐 3D 高斯模型进行人体头发重建
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
[**论文**](https://arxiv.org/abs/2409.14778) | [**项目页面**](https://eth-ait.github.io/GaussianHaircut/)
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
本仓库包含了 Gaussian Haircut 的官方实现,这是一种基于股线的人体头发重建方法,用于单目视频。
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 03:35:06 +08:00
|
|
|
|
## 预览
|
|
|
|
|
|
|
|
|
|
重建过程包括以下主要阶段:
|
|
|
|
|
|
|
|
|
|
1. **预处理阶段**
|
|
|
|
|
- 视频帧提取和整理
|
|
|
|
|
- COLMAP相机重建
|
|
|
|
|
- 头发和身体分割
|
|
|
|
|
- 图像质量评估和筛选
|
|
|
|
|
- 方向图计算
|
|
|
|
|
- 人脸关键点检测
|
|
|
|
|
- FLAME头部模型拟合
|
|
|
|
|
|
|
|
|
|
2. **重建阶段**
|
|
|
|
|
- 3D高斯体重建
|
|
|
|
|
- FLAME网格拟合
|
|
|
|
|
- 场景裁剪和优化
|
|
|
|
|
- 头发股线重建
|
|
|
|
|
|
|
|
|
|
3. **可视化阶段**
|
|
|
|
|
- 导出重建的股线
|
|
|
|
|
- Blender渲染可视化
|
|
|
|
|
- 生成结果视频
|
|
|
|
|
|
|
|
|
|
预期输出:
|
|
|
|
|
```
|
|
|
|
|
[your_scene_folder]/
|
|
|
|
|
├── raw.mp4 # 输入视频
|
|
|
|
|
├── 3d_gaussian_splatting/ # 3D高斯体重建结果
|
|
|
|
|
├── flame_fitting/ # FLAME头部模型拟合结果
|
|
|
|
|
├── strands_reconstruction/ # 头发股线重建中间结果
|
|
|
|
|
├── curves_reconstruction/ # 最终头发股线结果
|
|
|
|
|
└── visualization/ # 渲染结果和视频
|
|
|
|
|
```
|
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
## 入门指南
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
### Linux 平台
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
1. **安装 CUDA 11.8**
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
按照 https://developer.nvidia.com/cuda-11-8-0-download-archive 上的说明进行操作。
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
确保:
|
|
|
|
|
- PATH 包含 <CUDA_DIR>/bin
|
|
|
|
|
- LD_LIBRARY_PATH 包含 <CUDA_DIR>/lib64
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
该环境仅在此 CUDA 版本下进行了测试。
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
2. **安装 Blender 3.6** 以创建股线可视化
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
按照 https://www.blender.org/download/lts/3-6 上的说明进行操作。
|
|
|
|
|
|
|
|
|
|
3. **克隆仓库并运行安装脚本**
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
|
|
|
|
```bash
|
|
|
|
|
git clone git@github.com:eth-ait/GaussianHaircut.git
|
|
|
|
|
cd GaussianHaircut
|
|
|
|
|
chmod +x ./install.sh
|
|
|
|
|
./install.sh
|
|
|
|
|
```
|
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
### Windows 平台
|
|
|
|
|
|
|
|
|
|
1. **安装 CUDA 11.8**
|
|
|
|
|
- 从 https://developer.nvidia.com/cuda-11-8-0-download-archive 下载并安装
|
|
|
|
|
- 默认安装路径:C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
|
2025-02-15 14:31:02 +08:00
|
|
|
|
- 确保CUDA版本与系统兼容
|
2025-02-14 01:50:52 +08:00
|
|
|
|
|
|
|
|
|
2. **安装 Blender 3.6**
|
|
|
|
|
- 从 https://www.blender.org/download/lts/3-6 下载并安装
|
|
|
|
|
- 默认安装路径:C:\Program Files\Blender Foundation\Blender 3.6
|
|
|
|
|
|
|
|
|
|
3. **安装 Visual Studio 2019 Build Tools**
|
|
|
|
|
- 从 https://visualstudio.microsoft.com/vs/older-downloads/ 下载并安装
|
|
|
|
|
- 选择"C++构建工具"工作负载
|
|
|
|
|
- 默认安装路径:C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools
|
|
|
|
|
|
|
|
|
|
4. **安装 COLMAP**
|
|
|
|
|
- 从 https://github.com/colmap/colmap/releases 下载并安装
|
2025-02-15 14:31:02 +08:00
|
|
|
|
- 下载CUDA版本的COLMAP (例如:COLMAP-3.8-windows-cuda.zip)
|
|
|
|
|
- 解压到不含空格的路径 (例如:C:\COLMAP)
|
|
|
|
|
- 将COLMAP目录添加到系统PATH:
|
2025-02-14 02:42:40 +08:00
|
|
|
|
1. 打开"系统属性" > "环境变量"
|
|
|
|
|
2. 在"系统变量"中找到"Path"
|
|
|
|
|
3. 点击"编辑" > "新建"
|
2025-02-15 14:31:02 +08:00
|
|
|
|
4. 添加COLMAP目录路径
|
|
|
|
|
5. 点击"确定"保存
|
|
|
|
|
- 重启终端使PATH生效
|
2025-02-14 01:50:52 +08:00
|
|
|
|
|
2025-02-15 16:08:48 +08:00
|
|
|
|
5. **安装 7-Zip**
|
|
|
|
|
- 从 https://7-zip.org/ 下载并安装
|
|
|
|
|
- 使用默认安装路径安装:
|
|
|
|
|
* 64位系统: C:\Program Files\7-Zip
|
|
|
|
|
* 32位系统: C:\Program Files (x86)\7-Zip
|
|
|
|
|
- 脚本会自动检测这些路径
|
2025-02-14 01:50:52 +08:00
|
|
|
|
|
2025-02-15 14:31:02 +08:00
|
|
|
|
6. **下载预训练模型和资源**
|
|
|
|
|
```cmd
|
|
|
|
|
# 在PowerShell中运行:
|
2025-02-15 16:08:48 +08:00
|
|
|
|
# 脚本会自动安装gdown并下载所需资源
|
2025-02-15 14:31:02 +08:00
|
|
|
|
.\download_resource.bat
|
|
|
|
|
```
|
2025-02-15 16:08:48 +08:00
|
|
|
|
注意:
|
|
|
|
|
- 下载过程可能需要几分钟到几十分钟,取决于网络速度
|
|
|
|
|
- 如果下载失败,可以重新运行脚本
|
|
|
|
|
- 确保有稳定的网络连接
|
2025-02-14 01:50:52 +08:00
|
|
|
|
|
2025-02-15 16:08:48 +08:00
|
|
|
|
6. **克隆仓库并运行安装脚本**
|
2025-02-14 01:50:52 +08:00
|
|
|
|
```cmd
|
2025-02-15 14:37:19 +08:00
|
|
|
|
git clone https://github.com/Jeffreytsai1004/GaussianHairCut
|
|
|
|
|
cd GaussianHairCut
|
|
|
|
|
# 先下载所需要的资源
|
|
|
|
|
.\download_resource.bat
|
|
|
|
|
# 运行安装脚本
|
2025-02-14 02:42:40 +08:00
|
|
|
|
.\install.bat
|
2025-02-15 14:37:19 +08:00
|
|
|
|
# 运行重建脚本
|
|
|
|
|
.\run.bat
|
2025-02-14 01:50:52 +08:00
|
|
|
|
```
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
## 重建
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
1. **录制单目视频**
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 02:42:40 +08:00
|
|
|
|
- 参考项目页面上的示例视频
|
|
|
|
|
- 录制要求:
|
|
|
|
|
* 拍摄对象应缓慢转动头部,确保捕捉到所有角度
|
|
|
|
|
* 保持头发和面部清晰可见
|
|
|
|
|
* 避免快速移动导致的运动模糊
|
|
|
|
|
* 保持光照条件稳定
|
|
|
|
|
* 建议视频长度:10-20秒
|
|
|
|
|
* 建议分辨率:1920x1080或更高
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
2. **设置重建场景的目录**
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 02:42:40 +08:00
|
|
|
|
创建场景目录结构:
|
|
|
|
|
```
|
|
|
|
|
[your_scene_folder]/
|
|
|
|
|
└── raw.mp4 # 您录制的视频文件
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
注意:
|
|
|
|
|
- DATA_PATH 应指向包含 raw.mp4 的目录
|
|
|
|
|
- 目录路径不应包含空格或特殊字符
|
|
|
|
|
- 脚本会在此目录下自动创建所需的子目录
|
|
|
|
|
- 确保有足够的磁盘空间(建议至少20GB)
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
3. **运行脚本**
|
|
|
|
|
|
|
|
|
|
Linux:
|
2025-02-14 00:44:47 +08:00
|
|
|
|
```bash
|
|
|
|
|
export PROJECT_DIR="[/path/to/]GaussianHaircut"
|
|
|
|
|
export BLENDER_DIR="[/path/to/blender/folder/]blender"
|
|
|
|
|
DATA_PATH="[path/to/scene/folder]" ./run.sh
|
|
|
|
|
```
|
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
Windows:
|
|
|
|
|
```cmd
|
2025-02-14 02:42:40 +08:00
|
|
|
|
# 在CMD中运行:
|
2025-02-14 01:50:52 +08:00
|
|
|
|
set PROJECT_DIR=[path\to\]GaussianHaircut
|
|
|
|
|
set DATA_PATH=[path\to\scene\folder]
|
|
|
|
|
run.bat
|
2025-02-14 02:19:55 +08:00
|
|
|
|
|
|
|
|
|
# 或在PowerShell中运行:
|
|
|
|
|
set PROJECT_DIR=[path\to\]GaussianHaircut
|
|
|
|
|
set DATA_PATH=[path\to\scene\folder]
|
|
|
|
|
.\run.bat
|
2025-02-14 01:50:52 +08:00
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
该脚本执行数据预处理、重建以及使用 Blender 进行最终可视化。使用 Tensorboard 查看中间可视化结果。
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
## 许可证
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
此代码基于 3D Gaussian Splatting 项目。有关条款和条件,请参阅 LICENSE_3DGS。其余代码根据 CC BY-NC-SA 4.0 分发。
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
如果此代码对您的项目有帮助,请引用以下论文。
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
## 引用
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
|
|
|
|
```
|
|
|
|
|
@inproceedings{zakharov2024gh,
|
|
|
|
|
title = {Human Hair Reconstruction with Strand-Aligned 3D Gaussians},
|
|
|
|
|
author = {Zakharov, Egor and Sklyarova, Vanessa and Black, Michael J and Nam, Giljoo and Thies, Justus and Hilliges, Otmar},
|
|
|
|
|
booktitle = {European Conference of Computer Vision (ECCV)},
|
|
|
|
|
year = {2024}
|
|
|
|
|
}
|
|
|
|
|
```
|
|
|
|
|
|
2025-02-14 01:50:52 +08:00
|
|
|
|
## 链接
|
2025-02-14 00:44:47 +08:00
|
|
|
|
|
|
|
|
|
- [3D Gaussian Splatting](https://github.com/graphdeco-inria/gaussian-splatting)
|
2025-02-14 01:50:52 +08:00
|
|
|
|
- [Neural Haircut](https://github.com/SamsungLabs/NeuralHaircut): FLAME 拟合管线、股线先验和发型扩散先验
|
|
|
|
|
- [HAAR](https://github.com/Vanessik/HAAR): 头发上采样
|
|
|
|
|
- [Matte-Anything](https://github.com/hustvl/Matte-Anything): 头发和身体分割
|
|
|
|
|
- [PIXIE](https://github.com/yfeng95/PIXIE): FLAME 拟合的初始化
|
|
|
|
|
- [Face-Alignment](https://github.com/1adrianb/face-alignment), [OpenPose](https://github.com/CMU-Perceptual-Computing-Lab/openpose): 用于 FLAME 拟合的关键点检测
|