Install
$ agentstack add skill-ostrichhermit-oh-unity-gamedev-skills-media-pipe-unity-skill ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
MediaPipe Unity Plugin Skill
Google MediaPipe 在 Unity 中的完整集成,提供手部追踪、面部检测、姿态识别、手势识别、目标检测等计算机视觉和机器学习解决方案。
何时使用此 Skill
在以下情况下触发此 Skill :
- MediaPipe Unity Plugin (homuler/com.github.homuler.mediapipe)
- 计算机视觉任务 (Computer Vision Tasks)
- 手部关键点检测 (Hand Landmark Detection)
- 面部检测和关键点 (Face Detection & Landmarks)
- 姿态关键点检测 (Pose Landmark Detection)
- 手势识别 (Gesture Recognition)
- 目标检测 (Object Detection)
- 图像分割 (Image Segmentation)
- 音频分类 (Audio Classification)
- 自定义 Calculator 图 (Custom Calculator Graph)
- GPU 推理优化 (GPU Inference)
快速参考
Hello World - 基础示例
using Mediapipe;
using UnityEngine;
public sealed class HelloWorld : MonoBehaviour
{
private const string ConfigText = @"
input_stream: ""in""
output_stream: ""out""
node {
calculator: ""PassThroughCalculator""
input_stream: ""in""
output_stream: ""out1""
}
node {
calculator: ""PassThroughCalculator""
input_stream: ""out1""
output_stream: ""out""
}
";
private void Start()
{
using var graph = new CalculatorGraph(ConfigText);
using var poller = graph.AddOutputStreamPoller("out");
graph.StartRun();
for (var i = 0; i ();
while (poller.Next(packet))
{
Debug.Log(packet.Get());
}
graph.WaitUntilDone();
}
}
1. CalculatorGraph - 核心图管理
using Mediapipe;
using UnityEngine;
public class GraphExample : MonoBehaviour
{
private CalculatorGraph _graph;
void Start()
{
// 初始化图
var configText = @"
input_stream: ""input_video""
output_stream: ""output_video""
node {
calculator: ""PassThroughCalculator""
input_stream: ""input_video""
output_stream: ""output_video""
}
";
_graph = new CalculatorGraph(configText);
// 启用 GPU 推理(可选)
// var gpuResources = GpuResources.Create();
// _graph.SetGpuResources(gpuResources);
// 添加输出流监听器
_graph.StartRun();
}
void OnDestroy()
{
// 清理资源
_graph?.CloseAllPacketSources();
_graph?.WaitUntilDone();
_graph?.Dispose();
}
}
2. Task API - 现代化解决方案接口
using Mediapipe.Tasks.Vision.HandLandmarker;
using Mediapipe.Tasks.Vision.Core;
using UnityEngine;
public class HandDetectionExample : MonoBehaviour
{
private HandLandmarker _handLandmarker;
void Start()
{
// 创建 Hand Landmarker
var options = new HandLandmarkerOptions
{
BaseOptions = new BaseOptions(ModelAssetPath: "hand_landmarker.task"),
RunningMode = RunningMode.LIVE_STREAM,
NumHands = 2,
MinHandDetectionConfidence = 0.5f,
MinHandPresenceConfidence = 0.5f,
MinTrackingConfidence = 0.5f
};
_handLandmarker = HandLandmarker.CreateFromOptions(options);
}
void Update()
{
// 检测手部关键点
var image = new Mediapipe.Tasks.Vision.Image();
// ... 设置图像数据
var result = _handLandmarker.Detect(image);
}
void OnDestroy()
{
_handLandmarker?.Dispose();
}
}
3. Legacy Solution - 图运行器模式
using Mediapipe.Unity.Sample;
using UnityEngine;
public class SolutionRunner : GraphRunner
{
// 配置文件在 Inspector 中设置
// [SerializeField] private TextAsset _cpuConfig;
protected override void ConfigureCalculatorGraph(CalculatorGraphConfig config)
{
// 配置输出流
calculatorGraph.Initialize(config);
// 输出流设置...
}
public override void StartRun(ImageSource imageSource)
{
// 启动推理
StartRun(new PacketMap());
}
protected override IList RequestDependentAssets()
{
// 加载模型文件
return new List
{
WaitForAsset("model.binarypb")
};
}
}
4. 图像输入处理
using Mediapipe;
using Mediapipe.Unity.Sample;
using UnityEngine;
public class ImageInputExample : MonoBehaviour
{
private ImageSource _imageSource;
void Start()
{
// 使用 WebCam 作为输入源
_imageSource = ImageSourceProvider.ImageSource;
StartCoroutine(_imageSource.Play());
}
void Update()
{
if (!_imageSource.isPrepared) return;
// 获取当前帧
var currentTexture = _imageSource.GetCurrentTexture();
// 创建 ImageFrame
var imageFrame = new ImageFrame(
ImageFormat.Types.Format.SRGBA,
currentTexture.width,
currentTexture.height,
currentTexture.GetRawTextureData(),
ImageFrame.Types.GlFormatInfo.ForUnityTextureFormat(currentTexture.format)
);
// 发送到 CalculatorGraph
// graph.AddPacketToInputStream("input_video", Packet.CreateImageFrameAt(imageFrame, timestamp));
}
}
5. 处理输出结果
using Mediapipe;
using UnityEngine;
public class OutputHandler : MonoBehaviour
{
// 使用 OutputStreamPoller 同步获取结果
void ProcessOutputSync()
{
var poller = _graph.AddOutputStreamPoller("landmarks");
while (true)
{
var packet = new Packet();
if (poller.Next(packet))
{
var landmarks = packet.Get();
// 处理关键点数据
}
}
}
// 使用 NativePacketCallback 异步获取结果
void ProcessOutputAsync()
{
_graph.ObserveOutputStream("landmarks", LandmarkCallback);
[AOT.MonoPInvokeCallback(typeof(CalculatorGraph.NativePacketCallback))]
private static IntPtr LandmarkCallback(IntPtr graphPtr, int streamId, IntPtr packetPtr)
{
using (var packet = Packet.CreateForReference(packetPtr))
{
var landmarks = packet.Get();
// 处理关键点数据(注意:在非主线程执行)
}
return IntPtr.Zero;
}
}
}
支持的解决方案
视觉解决方案
| Solution | Description | Android | iOS | Linux | macOS | Windows | |----------|-------------|:-------:|:---:|:-----:|:-----:|:-------:| | Object Detection | 目标检测 | ✓ | ✓ | ✓ | ✓ | ✓ | | Face Detection | 面部检测 | ✓ | ✓ | ✓ | ✓ | ✓ | | Face Landmark Detection | 面部关键点检测 (468点) | ✓ | ✓ | ✓ | ✓ | ✓ | | Hand Landmark Detection | 手部关键点检测 (21点) | ✓ | ✓ | ✓ | ✓ | ✓ | | Gesture Recognition | 手势识别 | ✓ | ✓ | ✓ | ✓ | ✓ | | Pose Landmark Detection | 姿态关键点检测 (33点) | ✓ | ✓ | ✓ | ✓ | ✓ | | Image Segmentation | 图像分割 | ✓ | ✓ | ✓ | ✓ | ✓ | | Image Classification | 图像分类 | - | - | - | - | - | | Image Embedding | 图像嵌入 | - | - | - | - | - |
音频解决方案
| Solution | Description | Android | iOS | Linux | macOS | Windows | |----------|-------------|:-------:|:---:|:-----:|:-----:|:-------:| | Audio Classification | 音频分类 | ✓ | ✓ | ✓ | ✓ | ✓ |
传统解决方案
以下解决方案可通过 MediaPipe Framework 使用,但官方已停止支持:
- Holistic Tracking
- Face Mesh
- Hand Tracking
- Pose Tracking
- Objectron
- Instant Motion Tracking
API 概览
核心 API
CalculatorGraph
MediaPipe Framework 的主要 API,用于构建和运行计算图。
快速开始:
// 初始化
var graph = new CalculatorGraph(configText);
// GPU 模式
var gpuResources = GpuResources.Create();
graph.SetGpuResources(gpuResources);
// 运行
graph.StartRun();
graph.AddPacketToInputStream("input", packet);
graph.CloseInputStream("input");
graph.WaitUntilDone();
详细文档: [references/API-Overview.md](references/API-Overview.md#calculatorgraph)
Packet
MediaPipe 中数据传递的基本单位。
类型支持:
// 基本类型
Packet.CreateString("text")
Packet.CreateInt(42)
Packet.CreateBool(true)
Packet.CreateFloat(3.14f)
// 媒体类型
Packet.CreateImageFrame(imageFrame)
Packet.CreateGpuBuffer(gpuBuffer)
// 自定义时间戳
Packet.CreateStringAt("text", timestampMicrosec)
详细文档: [references/API-Overview.md](references/API-Overview.md#packett)
OutputStreamPoller
同步轮询输出流的数据。
// 创建轮询器
var poller = graph.AddOutputStreamPoller("landmarks");
// 获取数据(阻塞直到有新数据)
var packet = new Packet();
if (poller.Next(packet)) {
var data = packet.Get();
}
// 使用 observeTimestampBounds 处理空包
var poller = graph.AddOutputStreamPoller("landmarks", true);
详细文档: [references/API-Overview.md](references/API-Overview.md#outputstream)
任务 API
Task API 提供了更高级的接口,简化了常见任务的使用。
HandLandmarker
手部关键点检测和手势识别。
var options = new HandLandmarkerOptions {
BaseOptions = new BaseOptions(ModelAssetPath: "hand_landmarker.task"),
NumHands = 2,
MinHandDetectionConfidence = 0.5f
};
var landmarker = HandLandmarker.CreateFromOptions(options);
var result = landmarker.Detect(image);
// 访问结果
foreach (var landmarks in result.Landmarks) {
foreach (var landmark in landmarks) {
Debug.Log($"({landmark.X}, {landmark.Y}, {landmark.Z})");
}
}
FaceDetector / FaceLandmarker
面部检测和关键点识别。
// Face Detection
var detector = FaceDetector.CreateFromOptions(options);
var detectionResult = detector.Detect(image);
// Face Landmark Detection
var landmarker = FaceLandmarker.CreateFromOptions(options);
var landmarkResult = landmarker.Detect(image);
PoseLandmarker
姿态关键点检测。
var landmarker = PoseLandmarker.CreateFromOptions(options);
var result = landmarker.Detect(image);
// 33 个姿态关键点
// PoseLandmark 类定义了所有关键点索引
更多 Task API: [references/Tutorial-Task-API.md](references/Tutorial-Task-API.md)
Unity 特定 API
ImageSource
图像输入源管理(Webcam、视频文件、静态图片)。
// 获取图像源
var imageSource = ImageSourceProvider.ImageSource;
// 配置图像源
imageSource.SourceType = ImageSourceType.WebCamera;
imageSource.WebCameraDeviceIndex = 0;
// 启动
yield return imageSource.Play();
// 获取当前帧
var texture = imageSource.GetCurrentTexture();
GraphRunner
Unity 中 CalculatorGraph 的抽象基类,简化了图的生命周期管理。
public class MyGraphRunner : GraphRunner
{
protected override void ConfigureCalculatorGraph(CalculatorGraphConfig config)
{
// 配置图
calculatorGraph.Initialize(config);
}
public override void StartRun(ImageSource imageSource)
{
// 启动推理
}
protected override IList RequestDependentAssets()
{
// 加载模型和资源
return new List {
WaitForAsset("model.binarypb")
};
}
}
参考文件
此 Skill 包含来自 homuler/MediaPipeUnityPlugin 的完整文档,位于 references/ 目录:
入门指南
- [Getting-Started.md](references/Getting-Started.md) - 构建、安装和测试完整指南
- [Tutorial-Hello-World.md](references/Tutorial-Hello-World.md) - Hello World 教程
- [Tutorial-Task-API.md](references/Tutorial-Task-API.md) - Task API 使用教程
API 文档
- [API-Overview.md](references/API-Overview.md) - 核心 API 概览 (CalculatorGraph, Packet, GpuResources, etc.)
高级主题
- [Advanced-Topics.md](references/Advanced-Topics.md) - 高级用法和自定义计算器
- [Installation-Guide.md](references/Installation-Guide.md) - 详细安装指南
- [FAQ.md](references/FAQ.md) - 常见问题解答
项目 README
- [MediaPipeUnityPlugin-README.md](references/MediaPipeUnityPlugin-README.md) - 项目 README 和示例说明
使用此 Skill
初学者
- 阅读 [references/Tutorial-Hello-World.md](references/Tutorial-Hello-World.md) 了解基本概念
- 运行项目中的示例场景(
Assets/MediaPipeUnity/Samples/Scenes/) - 查看 Hello World 示例(
Assets/MediaPipeUnity/Tutorial/Hello World/HelloWorld.cs)
常见任务
- 手部追踪: 查看
HandLandmarkerRunner.cs和 Hand Landmark Detection 场景 - 面部检测: 查看
FaceDetectorRunner.cs和 Face Detection 场景 - 姿态识别: 查看
PoseLandmarkerRunner.cs和 Pose Landmark Detection 场景 - 目标检测: 查看
ObjectDetectorRunner.cs和 Object Detection 场景
高级用户
- 研究 [references/API-Overview.md](references/API-Overview.md) 了解底层 API
- 阅读 [references/Advanced-Topics.md](references/Advanced-Topics.md) 学习自定义 Calculator
- 参考官方 MediaPipe 文档: https://ai.google.dev/edge/mediapipe
平台支持
| Platform | CPU | GPU | Notes | |----------|:---:|:---:|-------| | Linux (x8664) | ✓ | ✓ | 推荐用于开发 | | macOS (Intel) | ✓ | ✗ | GPU 不支持 | | macOS (ARM64) | ✓ | ✗ | GPU 不支持 | | Windows (x8664) | ✓ | ✗ | GPU 实验性支持 | | Android | ✓ | ✓ | 需要 OpenGL ES 3.0+ | | iOS | ✓ | ✓ | 需要 Metal | | WebGL | ✗ | ✗ | 不支持 |
> 注意: GPU 模式在 macOS 和 Windows 上不支持。Windows 上 GPU 推理为实验性功能。
技术限制
UnityEditor 可能崩溃
由于 MediaPipe 使用原生库,某些错误可能导致 UnityEditor 崩溃(尤其是在 Windows 上)。
- Linux 和 macOS 上插件处理了 SIGABRT 信号以避免崩溃
- Windows 上无法正确处理 SIGABRT
Graphics API
使用 GPU 推理时,不能使用 OpenGL Core API,否则会出现错误:
InternalException: INTERNAL: ; eglMakeCurrent() returned error 0x3000
解决方案: 在 PC Standalone 构建中切换到 Vulkan。
模型文件位置
- UnityEditor: 使用 Local 模式(默认)
- 设备构建: 使用 StreamingAssets 模式,需要将模型文件复制到
StreamingAssets/目录
Android 构建注意事项
需要确保 APK 中包含 libstdc++_shared.so,否则会抛出 DllNotFoundException。
核心概念
MediaPipe 架构
- CalculatorGraph: 计算图,连接多个 Calculator 节点
- Calculator: 计算节点,处理输入并产生输出
- Packet: 数据包,在图中传递数据的基本单位
- InputStream / OutputStream: 输入和输出流
- SidePacket: 侧包,用于传递配置参数
图配置
使用文本格式配置计算图:
input_stream: "input_video"
output_stream: "output_video"
node {
calculator: "PassThroughCalculator"
input_stream: "input_video"
output_stream: "output_video"
}
坐标系统
- Unity: 左下角为原点 (0, 0)
- MediaPipe: 右上角为原点 (0, 0)
- 需要进行坐标转换(参见
CoordinateSystem类)
推理模式
- CPU: 使用 CPU 计算,兼容性好
- GPU: 使用 GPU 加速,性能更好(在支持的平台上)
- OpenGL ES: Android 平台优化的 GPU 模式
资源
官方资源
- MediaPipe GitHub: https://github.com/google/mediapipe
- MediaPipe Unity Plugin: https://github.com/homuler/MediaPipeUnityPlugin
- MediaPipe 官方文档: https://ai.google.dev/edge/mediapipe
- MediaPipe Concepts: https://ai.google.dev/edge/mediapipe/framework/framework_concepts/overview
示例场景位置
示例场景位于项目中的 Assets/MediaPipeUnity/Samples/Scenes/:
Face Detection/Face Landmark Detection/Hand Landmark Detection/Pose Landmark Detection/Object Detection/Image Segmentation/Legacy/(传统解决方案)
常用命名空间
using Mediapipe; // 核心 API
using Mediapipe.Tasks.Vision; // 视觉任务
using Mediapipe.Tasks.Vision.HandLandmarker; // 手部检测
using Mediapipe.Unity.Sample; // 示例代码
scripts/
帮助脚本和自动化工具可放在此处。
assets/
模板、样板代码或示例 Unity 项目文件可放在此处。
说明
- 此 Skill 基于 MediaPipe Unity Plugin v0.10.22 (homuler)
- 支持 Unity 2022.3+
- 参考文档来自官方 GitHub Wiki
- Task API 是推荐的使用方式,Legacy Solution 仅用于兼容
- 首次使用建议运行示例场景验证安装
此 Skill 基于 homuler/MediaPipeUnityPlugin 的官方文档和示例生成。
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: OstrichHermit
- Source: OstrichHermit/OH-Unity-GameDev-Skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.