<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/"><channel><title>XiYang</title><description>A personal site for projects, technical notes, topic maps, and structured documentation.</description><link>https://xiyang.work/</link><language>en</language><copyright>Copyright © 2026 Qi Yang</copyright><managingEditor>Qi Yang</managingEditor><webMaster>Qi Yang</webMaster><lastBuildDate>Mon, 01 Jun 2026 03:20:28 GMT</lastBuildDate><generator>Astro RSS</generator><docs>https://www.rssboard.org/rss-specification</docs><ttl>60</ttl><item><title>FaceFusion人脸替换与增强平台</title><link>https://xiyang.work/posts/facefusion人脸替换与增强平台/</link><guid isPermaLink="true">https://xiyang.work/posts/facefusion人脸替换与增强平台/</guid><description>FaceFusion Industry leading face manipulation platform — 开源人脸替换与增强平台，支持图片和视频处理。 概述 FaceFusion 是一个功能丰富的人脸操作平台，提供人脸替换（face swapping）、人脸增强（face enhancement）、表情编辑、年</description><pubDate>Fri, 29 May 2026 00:00:00 GMT</pubDate><author>Qi Yang</author></item><item><title>知识库索引</title><link>https://xiyang.work/posts/knowledge-base/</link><guid isPermaLink="true">https://xiyang.work/posts/knowledge-base/</guid><description>Personal knowledge catalog for the Hermes profile. Last updated: 2026 05 28 | Total pages: 52 Entities AI导览故事生成工具 — AI 导览故事生成工具：Gradio 应用 + 5 阶段技能管线，博物馆展品趣味文案自动</description><pubDate>Fri, 29 May 2026 00:00:00 GMT</pubDate><author>Qi Yang</author></item><item><title>腾讯混元WorldMirror重建模型</title><link>https://xiyang.work/posts/腾讯混元worldmirror重建模型/</link><guid isPermaLink="true">https://xiyang.work/posts/腾讯混元worldmirror重建模型/</guid><description>HunyuanWorld Mirror 腾讯混元团队开发的多功能前馈 3D 世界重建模型，单次前向传播即可同时生成多种 3D 表示。 概述 HunyuanWorld Mirror（论文名 WorldMirror）是一个统一的前馈神经网络，用于全面的 3D 几何预测。它整合多种几何先验（相机位姿、校准内参、深度图），在单</description><pubDate>Fri, 29 May 2026 00:00:00 GMT</pubDate><author>Qi Yang</author></item><item><title>AIGroupChat 剧本生成系统</title><link>https://xiyang.work/posts/aigroupchat剧本生成系统/</link><guid isPermaLink="true">https://xiyang.work/posts/aigroupchat剧本生成系统/</guid><description>该系统负责把单活动讲解点数据、机器人模板和运营配置转成结构化群聊剧本，当前主输出版本为 。 系统组成 1. 生成入口 主要接口 : 主要输入 : 或直接传入 （技术验证时可直接注入活动快照） 2. 数据准备 关键函数 : ：决定使用管理端实时读取还是 inline 快照 ：归一化活动、展馆、展厅、讲解点数据 / ：压缩</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>FastAPI</category><category>导览</category><category>AI</category><author>Qi Yang</author></item><item><title>AIGroupChat 源数据与存储系统</title><link>https://xiyang.work/posts/aigroupchat源数据与存储系统/</link><guid isPermaLink="true">https://xiyang.work/posts/aigroupchat源数据与存储系统/</guid><description>该系统负责从管理端读取活动/讲解点数据，并以文件形式持久化机器人模板与静态剧本，是 AIGroupChat 技术验证阶段的数据底座。 系统组成 1. 管理端源数据读取 关键服务 : 行为 : 读取点亮艺术管理端活动及讲解点接口 支持按 限定单活动 支持固定 token 或账号密码自动登录 遇到 401/403 时清空内</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>FastAPI</category><author>Qi Yang</author></item><item><title>AIGroupChat</title><link>https://xiyang.work/posts/aigroupchat群聊导览剧本系统/</link><guid isPermaLink="true">https://xiyang.work/posts/aigroupchat群聊导览剧本系统/</guid><description>AIGroupChat 是一个面向“单活动 AI 群聊导览剧本”的技术验证项目。它把管理端活动/讲解点数据、机器人模板、静态剧本生成、剧本发布校验，以及运行时用户插话回复串成一个完整闭环。 项目概述 后端框架 : FastAPI 核心协议 : Responses API 数据边界 : 一个群聊实例只绑定一个活动，剧本中</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>FastAPI</category><category>导览</category><category>AI</category><author>Qi Yang</author></item><item><title>AIGroupChat 运行时回复系统</title><link>https://xiyang.work/posts/aigroupchat运行时回复系统/</link><guid isPermaLink="true">https://xiyang.work/posts/aigroupchat运行时回复系统/</guid><description>该系统负责在静态剧本播放过程中处理用户插话，并以“托住 + 绕回”的方式回复，同时保持主剧情继续推进，不破坏必推讲解点顺序。 系统组成 1. 运行时直接回复 接口 : 输入重点 : 边界要求 : 只能包含“已经推出”的 group/messages 不允许把未来尚未展示的内容传入模型 2. 托住并绕回 接口 : 用途 </description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>FastAPI</category><category>导览</category><author>Qi Yang</author></item><item><title>LightUP_AItourPoint_API</title><link>https://xiyang.work/posts/lightup讲解点ai生成api/</link><guid isPermaLink="true">https://xiyang.work/posts/lightup讲解点ai生成api/</guid><description>LightUP AItourPoint API 是一个面向讲解点语音生成的后端服务，围绕 这条链路组织接口与服务。 项目概述 后端框架 : FastAPI 接口前缀 : 核心外部能力 : Responses API 文本生成、StepFun TTS、Whisper/LRC 对齐 输出目标 : 讲解文稿、朗读要求、TTS</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>FastAPI</category><category>TTS</category><category>导览</category><category>AI</category><author>Qi Yang</author></item><item><title>LightUP 讲解点 AI 生成链路</title><link>https://xiyang.work/posts/lightup讲解点ai生成链路/</link><guid isPermaLink="true">https://xiyang.work/posts/lightup讲解点ai生成链路/</guid><description>该系统负责把讲解点上下文逐步加工为可用于语音合成的内容，链路为： 。 系统组成 1. 讲解文稿生成 接口 : 输入重点 : 校验规则 : 缺少 HTTP 400 但没有 HTTP 400 为空 HTTP 400 当前不支持 HTTP 400 2. 朗读要求生成 接口 : 用途 : 根据展品名称、作者、讲解文稿、图片等上</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>FastAPI</category><category>导览</category><category>TTS</category><category>AI</category><author>Qi Yang</author></item><item><title>LightUP 音频生成与 LRC 对齐系统</title><link>https://xiyang.work/posts/lightup音频生成与lrc对齐系统/</link><guid isPermaLink="true">https://xiyang.work/posts/lightup音频生成与lrc对齐系统/</guid><description>该系统负责把 提交给 StepFun 生成音频，并在音频成功后异步启动 LRC 对齐任务，最终通过任务查询接口返回字幕结果。 系统组成 1. 默认配置与音色列表 接口 : 默认值来源 : 当前音色白名单 : — 默认女声 — 中医药馆长 2. 音频生成入口 接口 : 核心校验 : 必须提供 必须提供 必须提供 必须固定</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>FastAPI</category><category>TTS</category><category>Whisper</category><author>Qi Yang</author></item><item><title>暂存索引</title><link>https://xiyang.work/posts/staging-index/</link><guid isPermaLink="true">https://xiyang.work/posts/staging-index/</guid><description>该页用于汇总 相关的 intake、处理中内容与归档入口。 当前状态 当前仅作为 中的占位导航页 后续可扩展为：staging 来源分类、处理进度、归档去向、未整理项清单 相关页面 主题图谱</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><author>Qi Yang</author></item><item><title>VRDragon Multiplayer Unity 项目</title><link>https://xiyang.work/posts/vrdragon多人龙舟unity项目/</link><guid isPermaLink="true">https://xiyang.work/posts/vrdragon多人龙舟unity项目/</guid><description>VRDragon Multiplayer 是一个基于 Unity + Mirror 的 VR 龙舟多人竞速项目，主场景为 。系统面向 Windows 服务端与 Android PICO 客户端，围绕“等待 → 上船 → 准备 → 自动开赛 → 划桨竞速 → 排行榜 → 连续下一局”的自动比赛流程构建。 项目概述 引擎/</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>Unity</category><category>VR</category><category>Mirror</category><category>Pico</category><author>Qi Yang</author></item><item><title>VRDragon 结果展示与现场运营系统</title><link>https://xiyang.work/posts/vrdragon结果展示与现场运营系统/</link><guid isPermaLink="true">https://xiyang.work/posts/vrdragon结果展示与现场运营系统/</guid><description>该系统覆盖比赛冲线判定、成绩 JSON 保存、排行榜刷新、服务端看板、观赛相机切换、完赛气氛与强制复位，是 VRDragon Multiplayer 面向线下运营的关键支撑层。 系统组成 1. 冲线与完赛判定 主要文件 : , 规则 : 终点触发器只在服务端记录结果 防止同一条船重复记分 首船冲线只记录成绩和触发表现，</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>Unity</category><author>Qi Yang</author></item><item><title>VRDragon 网络与比赛流程系统</title><link>https://xiyang.work/posts/vrdragon网络与比赛流程系统/</link><guid isPermaLink="true">https://xiyang.work/posts/vrdragon网络与比赛流程系统/</guid><description>该系统负责 VRDragon Multiplayer 的联机角色选择、客户端连接、玩家状态同步、上船分配、准备/开赛状态机，以及赛后返回等待的完整自动赛流程。 系统组成 1. 网络启动入口 主要文件 : , , 核心职责 : 根据运行环境自动调用 或 区分原工程 Editor 与 ParrelSync 克隆 Edito</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>Unity</category><category>Mirror</category><category>Pico</category><category>VR</category><author>Qi Yang</author></item><item><title>VRDragon 船体运动与调参系统</title><link>https://xiyang.work/posts/vrdragon船体运动与调参系统/</link><guid isPermaLink="true">https://xiyang.work/posts/vrdragon船体运动与调参系统/</guid><description>该系统负责检测玩家划桨、统一驱动玩家船与 NPC 船的推进、限制船速与赛道方向，并通过服务端运行时面板即时调整运动手感。 系统组成 1. 划桨输入检测 主要文件 : 作用 : 检测船桨进入 标签碰撞体 对本地划桨设置 冷却 本地触水后发起 与 限制条件 : 只有本地玩家对象处理输入 服务端只有在 且玩家状态为 时才接受</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>Unity</category><category>Mirror</category><author>Qi Yang</author></item><item><title>Wikilinks机制</title><link>https://xiyang.work/posts/wikilinks机制/</link><guid isPermaLink="true">https://xiyang.work/posts/wikilinks机制/</guid><description>是 Obsidian 的内部链接机制，用于把知识库页面连接成可导航网络。 当前用途 作为 中对知识库交叉引用能力的占位说明页 后续可扩展为：命名规范、重定向策略、坏链排查方法、索引维护规则 相关页面 主题图谱 技术文档结构化写作</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>Obsidian</category><author>Qi Yang</author></item><item><title>CameraPoint 相机点管理系统</title><link>https://xiyang.work/posts/camerapoint相机点管理系统/</link><guid isPermaLink="true">https://xiyang.work/posts/camerapoint相机点管理系统/</guid><description>CameraPoint 是 UE5 项目中的相机点管理系统，用于在虚拟展厅中创建、管理、导出和加载相机点位数据。 系统组成 核心工具 1. ExportTool (227KB): 主要导出工具，将相机点数据导出为JSON格式 2. ExportTool 24 (226KB): 24帧版本的导出工具 3. ExportT</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>UE5</category><author>Qi Yang</author></item><item><title>Frame Blueprint 画框蓝图系统</title><link>https://xiyang.work/posts/frame-blueprint画框蓝图系统/</link><guid isPermaLink="true">https://xiyang.work/posts/frame-blueprint画框蓝图系统/</guid><description>Frame Blueprint 是 UE5 项目中的画框管理系统，用于在虚拟展厅中管理画框的材质、模型、动画和数据流。系统有两个版本，V2 是简化重构版。 系统版本 Frame Blueprint (V1) 第一代画框系统，包含完整的上下游数据流和材质管理。 核心组件： Frame Blueprint (195KB):</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>UE5</category><author>Qi Yang</author></item><item><title>GZUCM_VRVideo UI 系统</title><link>https://xiyang.work/posts/gzucm-vrvideo-ui系统/</link><guid isPermaLink="true">https://xiyang.work/posts/gzucm-vrvideo-ui系统/</guid><description>UI 系统是 GZUCM VRVideo 项目的用户界面管理核心，负责菜单显示、音频控制、媒体播放控制等 UI 相关功能。 系统组成 1. MenuUI（菜单 UI 系统） 文件 : 核心功能 : 管理菜单显示/隐藏 处理场景切换提示 控制 UI 状态重置 关键组件 : 状态管理 : 单例模式 : 使用 DontDes</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>VR</category><author>Qi Yang</author></item><item><title>GZUCM_VRVideo Unity 项目</title><link>https://xiyang.work/posts/gzucm-vrvideo-unity项目/</link><guid isPermaLink="true">https://xiyang.work/posts/gzucm-vrvideo-unity项目/</guid><description>GZUCM VRVideo 是基于 Unity 开发的 VR 视频播放项目，主要面向 Pico 设备，实现了媒体展示、传送系统、交互式菜单等功能。 项目概述 引擎版本 : Unity (URP 渲染管线) 目标平台 : Pico VR 设备 核心功能 : VR 视频/图片展示、传送系统、交互式菜单、媒体播放控制 主要依</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>VR</category><category>Pico</category><category>XR</category><author>Qi Yang</author></item><item><title>GZUCM_VRVideo 交互系统</title><link>https://xiyang.work/posts/gzucm-vrvideo交互系统/</link><guid isPermaLink="true">https://xiyang.work/posts/gzucm-vrvideo交互系统/</guid><description>交互系统是 GZUCM VRVideo 项目的用户输入处理核心，基于 Unity XR Interaction Toolkit 实现抓取、悬停、选择等 VR 交互功能。 系统组成 1. GrabAttachReturn（抓取返回系统） 文件 : 核心功能 : 记录物体原始位置和旋转 抓取时取消动画 释放时平滑返回原位 </description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>VR</category><category>XR</category><author>Qi Yang</author></item><item><title>GZUCM_VRVideo 传送系统</title><link>https://xiyang.work/posts/gzucm-vrvideo传送系统/</link><guid isPermaLink="true">https://xiyang.work/posts/gzucm-vrvideo传送系统/</guid><description>传送系统是 GZUCM VRVideo 项目的核心导航功能，通过基于标签的传送机制实现 VR 环境中的快速位置移动。 系统组成 1. TeleportManager（传送管理器） 文件 : 核心功能 : 管理传送目的地配置 处理传送确认流程 控制菜单显示/隐藏 集成输入系统 关键组件 : 2. TeleportDest</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>VR</category><category>XR</category><author>Qi Yang</author></item><item><title>GZUCM_VRVideo 媒体展示系统</title><link>https://xiyang.work/posts/gzucm-vrvideo媒体展示系统/</link><guid isPermaLink="true">https://xiyang.work/posts/gzucm-vrvideo媒体展示系统/</guid><description>媒体展示系统是 GZUCM VRVideo 项目的核心功能，通过数据驱动的方式实现 VR 环境中的图片和视频展示。 系统组成 1. 数据层 InteractableData（交互数据组件） 文件 : 核心功能 : 挂载在可交互物体上 定义媒体类型（图片/视频） 存储媒体 URL 列表 字段 : Interactable</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>VR</category><author>Qi Yang</author></item><item><title>Level Management 展厅场景管理系统</title><link>https://xiyang.work/posts/level-management展厅场景管理系统/</link><guid isPermaLink="true">https://xiyang.work/posts/level-management展厅场景管理系统/</guid><description>Level Management 是 UE5 项目中的展厅场景管理系统，用于管理虚拟展厅的布局、模型、材质和灯光。 系统组成 1. 展厅场景 G04 展厅 位置 : 模型文件 : D Side (18KB): 展厅侧面 DaPengChe Mat (26KB): 大篷车材质 HuaDuo Mat (25KB): 花朵材</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>UE5</category><category>Datasmith</category><author>Qi Yang</author></item><item><title>MR_Workshop Unity 项目</title><link>https://xiyang.work/posts/mr-workshop-unity项目/</link><guid isPermaLink="true">https://xiyang.work/posts/mr-workshop-unity项目/</guid><description>MR Workshop 是基于 Unity 开发的混合现实（MR）交互项目，主要面向 Pico 设备，实现了手势识别、区域触发、积木搭建等交互功能。 项目概述 引擎版本 : Unity (URP 渲染管线) 目标平台 : Pico MR 设备 核心功能 : 手势抓取、区域触发、积木生成与计数、蓝图搭建 主要依赖 : U</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>MR</category><category>XR</category><category>Pico</category><author>Qi Yang</author></item><item><title>MR Workshop 事件通信系统</title><link>https://xiyang.work/posts/mr-workshop事件通信系统/</link><guid isPermaLink="true">https://xiyang.work/posts/mr-workshop事件通信系统/</guid><description>事件通信系统是 MR Workshop 项目的核心架构组件，通过静态事件实现系统间解耦通信，支持计数更新、区域完成等关键事件。 系统组成 1. EventHandle（事件中心） 文件 : 核心功能 : 定义全局静态事件 提供事件触发方法 实现系统间解耦通信 事件列表 : 触发方法 : 2. CloneNamePool</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>MR</category><author>Qi Yang</author></item><item><title>MR Workshop 区域触发系统</title><link>https://xiyang.work/posts/mr-workshop区域触发系统/</link><guid isPermaLink="true">https://xiyang.work/posts/mr-workshop区域触发系统/</guid><description>区域触发系统是 MR Workshop 项目的核心游戏机制，通过检测物体在指定区域内的停留时间来触发游戏事件。 系统架构 核心组件 1. ZoneTrigger（区域触发器） 文件 : 核心功能 : 检测物体进入、停留、离开区域 计时器管理（物体停留指定时间触发完成） 完成状态管理 可视化进度反馈 状态机 : 关键字段</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>MR</category><author>Qi Yang</author></item><item><title>MR Workshop 手势交互系统</title><link>https://xiyang.work/posts/mr-workshop手势交互系统/</link><guid isPermaLink="true">https://xiyang.work/posts/mr-workshop手势交互系统/</guid><description>手势交互系统是 MR Workshop 项目的核心交互方式，通过 Pico 设备的手势识别功能实现无控制器的自然交互。 系统组成 1. HandPoseGrabber（手势抓取器） 文件 : 核心功能 : 监听手势识别事件（Gesture Performed/Ended） 在手势识别时自动抓取最近的可交互物体 支持手动</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>MR</category><category>Pico</category><category>XR</category><author>Qi Yang</author></item><item><title>MR Workshop 游戏逻辑系统</title><link>https://xiyang.work/posts/mr-workshop游戏逻辑系统/</link><guid isPermaLink="true">https://xiyang.work/posts/mr-workshop游戏逻辑系统/</guid><description>游戏逻辑系统是 MR Workshop 项目的核心玩法实现，包含积木生成、计数管理、蓝图搭建等游戏机制。 系统组成 1. GrabSpawn（抓取生成系统） 文件 : 核心功能 : 抓取物体时记录原始位置 释放时在原位生成替身 支持无限补货机制 工作流程 : 关键特性 : 无限补货 : 每次抓取都生成新替身 延迟启用 </description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>MR</category><author>Qi Yang</author></item><item><title>PanoRendering UE5 项目</title><link>https://xiyang.work/posts/panorendering-ue5项目/</link><guid isPermaLink="true">https://xiyang.work/posts/panorendering-ue5项目/</guid><description>PanoRendering 是基于 Unreal Engine 5.3 开发的全景相机渲染系统，主要用于虚拟展厅中的画框展示和全景相机路径管理。 项目概述 引擎版本 : UE5.3 主要功能 : 全景相机渲染、画框系统、展厅场景管理 核心插件 : Camera360v2、VaRest、MoviePipelineMask</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>UE5</category><author>Qi Yang</author></item><item><title>Sequence Automation 序列自动化系统</title><link>https://xiyang.work/posts/sequence-automation序列自动化系统/</link><guid isPermaLink="true">https://xiyang.work/posts/sequence-automation序列自动化系统/</guid><description>Sequence Automation 是 UE5 项目中的序列自动化系统，通过 Python 脚本从 JSON 配置文件自动生成 LevelSequence，实现相机路径的自动化创建。 系统组成 核心组件 1. CreateLevelSequenceFromTemplate.py : 主要脚本，从JSON配置创建Le</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>UE5</category><category>Python</category><author>Qi Yang</author></item><item><title>UE5 与 Unity MR 项目对比分析</title><link>https://xiyang.work/posts/ue5与unity-mr项目对比分析/</link><guid isPermaLink="true">https://xiyang.work/posts/ue5与unity-mr项目对比分析/</guid><description>本文对比分析两个 MR（混合现实）项目的架构设计、交互系统和实现方式。 项目概述 PanoRendering UE5 项目 引擎 : Unreal Engine 5.3 领域 : 全景相机渲染、展厅展示 核心 : 画框系统、相机路径管理 平台 : 通用 3D 应用 MR Workshop Unity 项目 引擎 : U</description><pubDate>Tue, 14 Apr 2026 00:00:00 GMT</pubDate><category>UE5</category><category>Unity</category><category>MR</category><category>XR</category><author>Qi Yang</author></item><item><title>3D摄影测量</title><link>https://xiyang.work/posts/3d摄影测量/</link><guid isPermaLink="true">https://xiyang.work/posts/3d摄影测量/</guid><description>3D 摄影测量与高斯溅射（Gaussian Splatting）相关的数据采集、格式转换与实时应用。 倾斜摄影（Oblique Photography） 倾斜摄影数据采集与输出： 使用多角度相机采集建筑物/场景的倾斜影像 常见输出格式：OSGB、OBJ、PLY Pix4D 等软件支持处理倾斜摄影数据 需查阅具体软件文档</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>3D</category><category>Unity</category><author>Qi Yang</author></item><item><title>ADB 多设备管理</title><link>https://xiyang.work/posts/adb多设备管理/</link><guid isPermaLink="true">https://xiyang.work/posts/adb多设备管理/</guid><description>当电脑同时连接多个 Android 设备或模拟器时， 命令会因无法确定目标设备而报错： 。 问题根源 在 MR/XR 开发中特别常见，原因包括： Android Studio 留下的 offline 模拟器残留 同时连接 Quest、Pico、手机等多台设备 Unity / Android Build Support </description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>XR</category><author>Qi Yang</author></item><item><title>AI图像生成工具链</title><link>https://xiyang.work/posts/ai图像生成工具链/</link><guid isPermaLink="true">https://xiyang.work/posts/ai图像生成工具链/</guid><description>AI 图像生成技术涵盖文本到图像、风格迁移、人脸替换、数字人等多个方向。本页整理相关工具链与实践经验。 ComfyUI 工作流核心组件 以 Flux 模型为例的 ComfyUI 核心节点： UNET Loader — 加载去噪网络模型 决定生成图像的基础能力和风格 Dual CLIP Loader — 双 CLIP 文</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>AI</category><category>ComfyUI</category><author>Qi Yang</author></item><item><title>AIGuidedTourStory</title><link>https://xiyang.work/posts/ai导览故事生成工具/</link><guid isPermaLink="true">https://xiyang.work/posts/ai导览故事生成工具/</guid><description>概述 AI Guided Tour Story 是一个面向博物馆/展览场景的 AI 导览故事生成工具。用户输入展品清单（Excel 表格或 JSON），系统通过 5 阶段技能管线自动输出趣味导览文案、主线故事和每件展品的趣味介绍终版。 当前冻结版本为 ，设计为局域网部署的 Gradio Web 应用。 技术栈 Pyth</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>Python</category><category>AI</category><category>导览</category><author>Qi Yang</author></item><item><title>AR 博物馆与手势交互方案</title><link>https://xiyang.work/posts/ar博物馆与手势交互方案/</link><guid isPermaLink="true">https://xiyang.work/posts/ar博物馆与手势交互方案/</guid><description>AR 在博物馆场景的落地方案，以及手势识别在 XR 交互中的应用。 AR 博物馆方案商 市场上提供 AR 博物馆整体解决方案的公司： 1. 云观博 ： 基于 AR 的智慧导览系统 已在近 500 家博物馆应用，包括近 50 家国家一级博物馆 2. 弥知科技（Kivicube） ： AR 制作平台，支持开发者制作 AR </description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>AR</category><category>导览</category><category>XR</category><category>ComfyUI</category><author>Qi Yang</author></item><item><title>CUDA Toolkit 安装与路径管理</title><link>https://xiyang.work/posts/cuda-toolkit安装与路径管理/</link><guid isPermaLink="true">https://xiyang.work/posts/cuda-toolkit安装与路径管理/</guid><description>核心结论 CUDA Toolkit 不必装在系统盘 ，可自由选择安装目录（如 ） 删除后可重新安装到其他盘，只要环境变量正确即可 CUDA 支持 多版本共存 ，通过切换 控制使用哪个版本 安装位置 默认路径： 这只是 NVIDIA 安装器的默认设定，并非强制要求。安装时选择 Custom Installation 即可</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>Windows</category><category>Configuration</category><author>Qi Yang</author></item><item><title>PyTorch</title><link>https://xiyang.work/posts/pytorch深度学习框架/</link><guid isPermaLink="true">https://xiyang.work/posts/pytorch深度学习框架/</guid><description>Stub page — 待补充详细内容 概述 PyTorch 是由 Meta（Facebook）开发的开源深度学习框架，广泛用于研究和生产环境。 核心组件 torch ：张量计算和自动微分 torch.nn ：神经网络模块 torch.optim ：优化器 torchvision / torchaudio / torc</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>AI</category><category>Python</category><author>Qi Yang</author></item><item><title>Qwen3TTS</title><link>https://xiyang.work/posts/qwen3-tts本地语音工具链/</link><guid isPermaLink="true">https://xiyang.work/posts/qwen3-tts本地语音工具链/</guid><description>本地部署的 Qwen3 TTS 语音合成 + Whisper 对齐工具链，面向博物馆讲解等场景的语音克隆与 LRC 歌词生成系统。 概览 Qwen3TTS 是一个基于 PyTorch深度学习框架 和 CUDA 的本地 TTS 工程项目，整合了阿里通义千问的 Qwen3 TTS 系列模型与 OpenAI Whisper </description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>TTS</category><category>AI</category><category>Whisper</category><category>Python</category><author>Qi Yang</author></item><item><title>主题图谱</title><link>https://xiyang.work/posts/topic-map/</link><guid isPermaLink="true">https://xiyang.work/posts/topic-map/</guid><description>高频主题导航页。随着 wiki 增长，把经常反复出现的主题聚合到这里，方便 Obsidian 浏览。 Last updated: 2026 05 28 XR/VR/AR 开发 ADB多设备管理 — 多设备 ADB 管理与设备选择 AR博物馆与手势交互方案 — AR 博物馆方案商、手势识别、ComfyUI AR 工作流 </description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><author>Qi Yang</author></item><item><title>UE5开发</title><link>https://xiyang.work/posts/ue5开发/</link><guid isPermaLink="true">https://xiyang.work/posts/ue5开发/</guid><description>Unreal Engine 5（UE5）开发涉及蓝图交互、编辑器扩展、资源导入、媒体播放等多个方向。本页汇总相关实践经验。 蓝图交互制作 在UE5中使用蓝图制作交互时，核心流程包括： 使用 FrameController 和 ParameterController 管理场景参数 基于 Plane 组件尺寸动态调整 li</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>UE5</category><category>Datasmith</category><author>Qi Yang</author></item><item><title>Unity SVN 文件清理</title><link>https://xiyang.work/posts/unity-svn文件清理/</link><guid isPermaLink="true">https://xiyang.work/posts/unity-svn文件清理/</guid><description>上传 Unity 工程到 SVN 时，以下文件/文件夹可以安全删除或在 中忽略： | 文件夹 | 原因 | | | | | Library | Unity 自动生成的编辑器缓存，可通过重新导入重建 | | Temp | 运行时临时文件，重启 Unity 会自动清除 | | obj | 编译时生成的中间文件 | | Bu</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><author>Qi Yang</author></item><item><title>Unity WebGL 开发</title><link>https://xiyang.work/posts/unity-webgl开发/</link><guid isPermaLink="true">https://xiyang.work/posts/unity-webgl开发/</guid><description>基于 Unity 的 WebGL 开发涉及跨平台构建、HTML 容器适配、视频播放策略、Shader 兼容性等多个技术维度。 Unity WebGL 项目核心技术栈 常见技术关键词： URP（Universal Render Pipeline）渲染管线 AVPro / VideoPlayer WebGL 视频播放（尤其</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>WebGL</category><category>Unity</category><category>3D</category><author>Qi Yang</author></item><item><title>Unity开发核心</title><link>https://xiyang.work/posts/unity开发核心/</link><guid isPermaLink="true">https://xiyang.work/posts/unity开发核心/</guid><description>Unity Development Core Unity 知识领域涵盖渲染管线、动画状态机、编辑器工具、网络编程、XR SDK 等。本页汇总个人 Unity 开发中的关键工具、模式与经验。 知识领域概述 Unity 开发涉及多个专业领域： 渲染管线 ：URP/Built in/自定义管线，Shader Graph，光照</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>XR</category><category>AI</category><author>Qi Yang</author></item><item><title>XR/VR/AR开发</title><link>https://xiyang.work/posts/xr-vr-ar开发/</link><guid isPermaLink="true">https://xiyang.work/posts/xr-vr-ar开发/</guid><description>XR/VR/AR Development XR（Extended Reality）开发涵盖 VR、AR、MR 技术，涉及 OpenXR 运行时、交互工具包、手势识别等领域。本页汇总个人 XR 开发经验与技术要点。 OpenXR 运行时 Monado OpenXR Monado 是开源 OpenXR 运行时，提供统一的 </description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>XR</category><category>VR</category><category>AR</category><author>Qi Yang</author></item><item><title>室内导航与定位技术</title><link>https://xiyang.work/posts/室内导航与定位技术/</link><guid isPermaLink="true">https://xiyang.work/posts/室内导航与定位技术/</guid><description>室内导航与定位技术预研，涵盖技术选型对比、硬件方案评估、成本预算等内容。 业务需求背景 室内导览系统需要解决的核心问题： 室内 GPS 信号不可用，需替代定位方案 游客在博物馆/展馆内的导航与讲解 技术选型需考虑精度、成本、可维护性 定位技术横向对比 从室内导览需求维度对比各定位技术（含具体参考数值）： | 维度 | </description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>导览</category><author>Qi Yang</author></item><item><title>技术文档结构化写作</title><link>https://xiyang.work/posts/技术文档结构化写作/</link><guid isPermaLink="true">https://xiyang.work/posts/技术文档结构化写作/</guid><description>技术文档的结构化写作方法，涵盖技术交底书、操作手册、流程图等常见文档类型。 技术交底书构成模块 技术交底书是专利申请前的核心文档，需包含以下模块： 1. 技术领域 — 所属技术领域概述 2. 背景技术 — 现有技术的不足与痛点 3. 发明内容 — 技术方案的整体描述 4. 技术方案 — 具体实现方式与创新点 5. 有益</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>Documentation</category><author>Qi Yang</author></item><item><title>杂项开发笔记</title><link>https://xiyang.work/posts/杂项开发笔记/</link><guid isPermaLink="true">https://xiyang.work/posts/杂项开发笔记/</guid><description>不属于主要主题集群的一次性对话和知识点汇总。每条记录保留核心要点，不展开长篇论述。 Shader 工作原理（Amplify Shader Editor） 使用 Amplify Shader Editor 时，需要先理解示例 shader 的输入/输出管线再改造： 区分 Vertex Shader（顶点变换）和 Frag</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>Unity</category><category>WebGL</category><author>Qi Yang</author></item><item><title>环境变量管理</title><link>https://xiyang.work/posts/环境变量管理/</link><guid isPermaLink="true">https://xiyang.work/posts/环境变量管理/</guid><description>Stub page — 待补充详细内容 概述 操作系统环境变量的配置、管理和多版本共存策略。 常见场景 PATH 管理 ：多个工具链版本共存时的路径优先级 LD LIBRARY PATH ：Linux 动态链接库搜索路径 用户级 vs 系统级 ： 、 、Windows 系统属性 相关页面 CUDA Toolkit安装与</description><pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate><category>Configuration</category><author>Qi Yang</author></item></channel></rss>