AutoMakeHuman

English

Overview

AutoMakeHuman is a MakeHuman plugin bundle for fitting a 3D human body or pose to a reference image.

This repository currently contains two plugins:

• 7_auto_human_body: adjusts body-related MakeHuman modifiers to better match a target silhouette
• 7_auto_human_pose: adjusts a BVH pose to better match a target pose image

What Was Modernized

The original project depended heavily on the author's local machine layout and older Python/OpenCV behavior. The current version has been upgraded so it is more practical to install and use today:

• plugin data paths are resolved relative to each plugin directory
• required data folders are created automatically
• OpenCV contour handling works with both OpenCV 3 and OpenCV 4
• several Python 3 incompatibilities were fixed, including rU, float slicing, and dict_values indexing
• image-processing utilities can be tested outside MakeHuman
• installation is scriptable on Windows, Linux, and macOS

Requirements

• Python 3.9 or newer
• MakeHuman with plugin support
• Python dependencies from requirements.txt

Install dependencies:

pip install -r requirements.txt

Install

Recommended: use the provided installer and point it to your MakeHuman plugins directory.

Cross-platform:

python bin/install.py --target /path/to/makehuman/plugins

Windows PowerShell:

.\bin\install.ps1 -Target "D:\MakeHuman\makehuman\plugins"

Linux/macOS:

bin/install.sh /path/to/makehuman/plugins

If MAKEHUMAN_PLUGIN_DIR is already set, the install scripts can use it directly.

Plugin Data

Each plugin reads and writes its own data under its package directory:

• src/7_auto_human_body/data
• src/7_auto_human_pose/data

Important bundled assets:

• body reference image: src/7_auto_human_body/data/MyAutoData/picture.jpg or picture.png
• pose reference image: src/7_auto_human_pose/data/MyAutoData/picture.png
• BVH pose files: src/7_auto_human_pose/data/MyAutoData/*.bvh

Usage

1. Install the plugins into the MakeHuman plugins directory.
2. Start or restart MakeHuman.
3. Open the Utilities category.
4. Use AutoBody to adjust body shape against a reference silhouette.
5. Use AutoPose to load or refine pose matching with the bundled BVH workflow.

Development

Basic validation:

python -m unittest discover -s test -p "test_*.py"

Optional debug windows for contour comparison:

AUTO_MAKEHUMAN_DEBUG_WINDOWS=1

Known Limits

• the pose plugin still contains a large amount of legacy search logic and repeated UI callback code
• the fitting strategy is still heuristic and not yet ML-based
• end-to-end validation still requires a real MakeHuman runtime

中文

项目简介

AutoMakeHuman 是一组用于 MakeHuman 的插件，目标是根据一张参考图片，自动拟合三维人体的体型或姿态。

当前仓库包含两个插件：

• 7_auto_human_body：通过调节 MakeHuman 的体型参数，让模型轮廓更接近目标图片
• 7_auto_human_pose：通过调整 BVH 姿态数据，让模型姿态更接近目标图片

已完成的现代化升级

原始项目强依赖作者本地目录结构，并且使用了较老版本的 Python 和 OpenCV 写法。当前版本已经做了第一轮可用性升级：

• 插件数据路径改为基于插件目录自动解析
• 运行所需的数据目录会自动创建
• 轮廓处理同时兼容 OpenCV 3 和 OpenCV 4
• 修复了多处 Python 3 兼容性问题，例如 rU、浮点切片、dict_values 下标访问
• 图像处理模块可以在 MakeHuman 外单独测试
• 安装流程支持 Windows、Linux 和 macOS

环境要求

• Python 3.9 及以上
• 支持插件的 MakeHuman
• requirements.txt 中列出的 Python 依赖

安装依赖：

pip install -r requirements.txt

安装方法

推荐使用仓库自带安装脚本，并将目标路径指向 MakeHuman 的 plugins 目录。

跨平台通用方式：

python bin/install.py --target /path/to/makehuman/plugins

Windows PowerShell：

.\bin\install.ps1 -Target "D:\MakeHuman\makehuman\plugins"

Linux/macOS：

bin/install.sh /path/to/makehuman/plugins

如果你已经设置了 MAKEHUMAN_PLUGIN_DIR 环境变量，安装脚本也可以直接使用它。

数据目录

两个插件都会在各自目录下读写数据：

• src/7_auto_human_body/data
• src/7_auto_human_pose/data

仓库中已包含的重要示例资源：

• 体型参考图：src/7_auto_human_body/data/MyAutoData/picture.jpg 或 picture.png
• 姿态参考图：src/7_auto_human_pose/data/MyAutoData/picture.png
• BVH 姿态文件：src/7_auto_human_pose/data/MyAutoData/*.bvh

使用说明

1. 将插件安装到 MakeHuman 的 plugins 目录
2. 启动或重启 MakeHuman
3. 打开 Utilities 分类
4. 在 AutoBody 中根据目标轮廓调整体型
5. 在 AutoPose 中加载或微调基于 BVH 的姿态匹配流程

开发与验证

基础验证命令：

python -m unittest discover -s test -p "test_*.py"

如果需要查看轮廓对比调试窗口，可以设置：

AUTO_MAKEHUMAN_DEBUG_WINDOWS=1

当前限制

• pose 插件内部仍保留较多历史搜索逻辑和重复的界面回调代码
• 当前自动拟合策略仍以启发式方法为主，还不是机器学习方案
• 完整的端到端验证仍然需要真实的 MakeHuman 运行环境