https://v5v5.fun/tags/%E8%AE%A1%E7%AE%97%E6%9C%BA%E8%A7%86%E8%A7%89/计算机视觉HugoBlox Kit (https://hugoblox.com)zh-CNSun, 01 Jun 2025 00:00:00 +0000https://v5v5.fun/media/sharing.pnghttps://v5v5.fun/tags/%E8%AE%A1%E7%AE%97%E6%9C%BA%E8%A7%86%E8%A7%89/https://v5v5.fun/projects/construction-safety-detection/Sun, 01 Jun 2025 00:00:00 +0000https://v5v5.fun/projects/construction-safety-detection/<p>面向<strong>工地安全管理</strong>的视觉检测系统，实时识别多种危险源与不安全行为，并将检测结果转化为可执行的预警与记录。</p> <h2 id="项目概述">项目概述</h2> <p>工地安全隐患类型多、环境多变：未佩戴安全帽或反光衣、周界入侵、人员跌倒、抽烟、烟火危险源等，仅靠人工盯屏难以覆盖。本项目围绕真实工地场景设计并验证了一套多目标视觉检测系统，覆盖从数据构建、模型研发到预警工作流的完整链路。</p> <h2 id="问题">问题</h2> <ul> <li>七类目标尺度与形态差异巨大——小目标烟火与大范围场景入侵共存，检测难度不同。</li> <li>工地光照、天气、机位角度多变，实验室指标无法直接代表现场效果。</li> <li>检测结果必须触发可执行的预警与记录，而非仅仅展示检测框。</li> <li>误报过多会让现场人员失去信任，漏报则带来安全风险，两者需共同控制。</li> </ul> <h2 id="解决方案">解决方案</h2> <p>设计 <strong>检测 → 预警 → 记录</strong> 的 AI 安全管控流程：</p> <ol> <li><strong>检测目标</strong>：安全帽、反光衣、周界入侵、人员跌倒、抽烟、人脸、烟火危险源七类目标。</li> <li><strong>模型研发</strong>：面向七类目标的检测模型设计、数据整理与增强、模型训练与效果评估。</li> <li><strong>场景适配</strong>：针对真实工地条件调优模型，并接入监控预警工作流。</li> </ol> <h2 id="系统架构">系统架构</h2> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-fallback" data-lang="fallback"><span class="line"><span class="cl">监控视频流 → 目标检测 → 多目标跟踪 → 行为分析 → 预警触发 → 记录归档 </span></span></code></pre></div><p> <figure > <div class="flex justify-center "> <div class="w-full" > <img alt="工地安全行为检测系统架构" srcset="https://v5v5.fun/projects/construction-safety-detection/architecture_hu_4651520429575699.webp 320w, https://v5v5.fun/projects/construction-safety-detection/architecture_hu_9c82be5ce728a174.webp 480w, https://v5v5.fun/projects/construction-safety-detection/architecture_hu_45f98379e86516d1.webp 760w" sizes="(max-width: 480px) 100vw, (max-width: 768px) 90vw, (max-width: 1024px) 80vw, 760px" src="https://v5v5.fun/projects/construction-safety-detection/architecture_hu_4651520429575699.webp" width="760" height="596" loading="lazy" data-zoomable /></div> </div></figure> </p> <h2 id="核心技术挑战">核心技术挑战</h2> <h3 id="复杂场景下的多类别检测">复杂场景下的多类别检测</h3> <p>七类目标在尺度、形态与出现频率上差异巨大，需要针对性的策略平衡精度：</p> <ul> <li>小目标（远处烟火）与大目标（周界入侵）采用差异化的处理策略平衡检测精度。</li> <li>数据整理与增强覆盖光照、天气与机位角度差异，缩小现场与训练分布的差距。</li> </ul> <h3 id="从检测到可执行的预警">从检测到可执行的预警</h3> <p>原始检测框不构成安全管理能力，还需要行为层面的加工：</p> <ul> <li>通过<strong>多目标跟踪</strong>关联时序信息，结合行为分析区分瞬时误检与持续风险。</li> <li>以持续帧数与区域规则约束触发条件，将<strong>误报 / 漏报</strong>压到现场可接受范围。</li> </ul> <h3 id="交付质量标准">交付质量标准</h3> <p>集成前建立明确的质量门槛，避免&quot;能跑即交付&quot;：</p> <ul> <li>在场景专属验证集上完成模型评估，制定清晰的精度与响应标准。</li> <li>场景适配迭代直至达标，再进入生产监控工作流。</li> </ul> <h2 id="工程实现">工程实现</h2> <ul> <li><strong>数据准备</strong>：标注整理、清洗与面向场景的增强策略。</li> <li><strong>模型研发</strong>：检测模型训练与迭代，在场景专属验证集上进行效果评估。</li> <li><strong>推理部署</strong>：基于 CUDA / ONNXRuntime 的推理优化，接入实时视频流。</li> <li><strong>业务集成</strong>：检测 + 跟踪 + 行为分析结果接入预警与记录工作流。</li> </ul> <h2 id="项目成果">项目成果</h2> <ul> <li><input checked="" disabled="" type="checkbox"> 七类危险源检测方案设计与验证全部完成</li> <li><input checked="" disabled="" type="checkbox"> 「检测 → 预警 → 记录」端到端安全管控流程落地</li> <li><input checked="" disabled="" type="checkbox"> 场景专属验证集与增强策略沉淀，支持后续迭代</li> <li><input checked="" disabled="" type="checkbox"> 模型在真实工地条件下完成场景适配并通过验收</li> </ul> <h2 id="演示">演示</h2> <p>摄像头画面实时检测安全隐患并产生预警与记录，覆盖全部七类目标。</p> <h2 id="技术栈">技术栈</h2> <p><strong>视觉模型</strong>：目标检测 · 多目标跟踪 · 行为分析 · PyTorch · CNN <strong>推理部署</strong>：CUDA · ONNXRuntime <strong>工程</strong>：数据标注与增强 · 效果评估 · 场景适配</p> <h2 id="我的贡献">我的贡献</h2> <p>负责算法设计与交付：</p> <ul> <li>七类目标的检测方案设计与验证</li> <li>数据整理、增强策略与场景专属验证集构建</li> <li>模型训练、评估与现场场景适配</li> <li>检测结果到预警记录工作流的集成落地</li> </ul> <p>该项目完整体现了我从 <strong>数据构建、检测模型训练与评估，到推理部署与安全管理流程落地</strong> 的计算机视觉工程能力。</p> <h2 id="应用前景与可扩展方向">应用前景与可扩展方向</h2> <p>当前「检测 → 预警 → 记录」流程已在真实工地场景跑通，同一框架具备进一步延展的空间：</p> <h3 id="应用场景">应用场景</h3> <ul> <li><strong>更多危险源类别</strong>：在同一框架下扩展未戴手套、违规用电等类别。</li> <li><strong>园区级规模接入</strong>：多路摄像头统一接入，形成片区级安全监测。</li> <li><strong>管理闭环联动</strong>：预警自动生成工单，与工地安全管理系统对接。</li> </ul> <h3 id="可扩展技术方向">可扩展技术方向</h3> <ul> <li><strong>边缘端部署</strong>：模型轻量化后下沉至边缘设备，降低带宽与延迟。</li> <li><strong>时序行为建模</strong>：引入时序模型识别更复杂的不安全行为模式。</li> <li><strong>主动干预提醒</strong>：结合现场音柱 / 屏幕做实时提示，形成&quot;发现即干预&quot;闭环。</li> </ul>https://v5v5.fun/projects/video-restoration/Sat, 01 Jun 2024 00:00:00 +0000https://v5v5.fun/projects/video-restoration/<div class="vr-wrap"> <h2 class="vr-sec"><span class="vr-sec-cn">项目概述</span><span class="vr-sec-en">Overview</span></h2> <p>长期从事计算机视觉与图像 / 视频处理相关算法研发，参与电影、电视剧、纪录片及历史影像的数字化修复与画质增强。</p> <p>技术方向从早期的图像降噪、缺陷修复和深度学习图像复原，逐步扩展到 CNN、Transformer、参考引导修复、超分辨率、时序一致性以及端到端视频复原。</p> <p>在算法研发之外，也参与数据集构建、模型评估、GPU 算力规划、批量任务处理和生产质量控制，将算法从实验环境落地到大规模视频处理场景。</p> <h2 class="vr-sec"><span class="vr-sec-cn">代表项目</span><span class="vr-sec-en">Featured Project</span></h2> <div class="vr-featured"> <p><span class="vr-featured-tag">Featured Project · 代表案例</span></p> <h3 id="黑龙江省广播电视局老旧剧集-4k-画质修复升级">黑龙江省广播电视局｜老旧剧集 4K 画质修复升级</h3> <p><strong>2023.12 – 2024.06</strong></p> <p>面向黑龙江省广播电视局历史节目资源开展老旧剧集 4K 超高清修复升级，对标实际播出与资源库建设需求，参与制定修复标准、处理流程及质量评价要求。</p> <p>针对历史视频中的噪声、模糊、细节缺失及色彩失真等问题，负责图像增强与视频修复算法研发，通过模型训练与参数优化持续改善修复效果；同时参与 GPU 资源规划、批量任务处理及成果质量控制。</p> <p>基于深度学习图像复原、Transformer、U-Net 及参考信息引导等技术开展算法研发，建立从数据处理、模型训练、批量 GPU 处理到最终质量验收的完整工程流程。</p> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-fallback" data-lang="fallback"><span class="line"><span class="cl">Historical Video </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Pre-processing </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Degradation Analysis </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Reference-guided Restoration </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Transformer + U-Net </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Super Resolution </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Temporal / Quality Control </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">GPU Batch Processing </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">4K UHD Output </span></span></code></pre></div><div class="grid gap-4 sm:grid-cols-2 lg:grid-cols-3"> <div class="rounded-2xl border border-gray-200 dark:border-white/10 bg-white dark:bg-white/[0.03] p-6 text-center h-full"> <div class="text-3xl sm:text-4xl font-bold text-primary-600 dark:text-primary-400">1,600&#43; min</div> <div class="mt-1.5 text-sm text-gray-600 dark:text-gray-400">历史视频处理量 · Historical Video Processed</div> </div> <div class="rounded-2xl border border-gray-200 dark:border-white/10 bg-white dark:bg-white/[0.03] p-6 text-center h-full"> <div class="text-3xl sm:text-4xl font-bold text-primary-600 dark:text-primary-400">4K UHD</div> <div class="mt-1.5 text-sm text-gray-600 dark:text-gray-400">最终输出分辨率 · Final Output Resolution</div> </div> <div class="rounded-2xl border border-gray-200 dark:border-white/10 bg-white dark:bg-white/[0.03] p-6 text-center h-full"> <div class="text-3xl sm:text-4xl font-bold text-primary-600 dark:text-primary-400">Accepted</div> <div class="mt-1.5 text-sm text-gray-600 dark:text-gray-400">项目通过正式验收 · Formal Acceptance</div> </div> </div> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">深度学习图像复原</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">Transformer</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">U-Net</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">参考信息引导</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">GPU 批量处理</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">质量控制</span> </div> </div> <h2 class="vr-sec"><span class="vr-sec-cn">技术演进</span><span class="vr-sec-en">Technical Evolution</span></h2> <div class="relative ml-2 pl-6 border-l-2 border-gray-200 dark:border-white/10 space-y-7"> <div class="relative"> <span class="absolute -left-[31px] top-1 w-3 h-3 rounded-full bg-primary-500 ring-4 ring-primary-500/20"></span> <div class="mb-1.5"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2015</span> <span class="ml-2 text-base font-bold text-gray-900 dark:text-white">电影 / 视频修复 Film &amp; Video Restoration</span> </div> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">图像降噪 · Image Denoising</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">画质增强 · Image Enhancement</span> </div> </div> <div class="relative"> <span class="absolute -left-[31px] top-1 w-3 h-3 rounded-full bg-primary-500 ring-4 ring-primary-500/20"></span> <div class="mb-1.5"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2017 – 2021</span> <span class="ml-2 text-base font-bold text-gray-900 dark:text-white">深度学习修复 Deep Learning Restoration</span> </div> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">CNN</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">视频去噪 · Video Denoising</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">超分辨率 · Super Resolution</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">时序处理 · Temporal Processing</span> </div> </div> <div class="relative"> <span class="absolute -left-[31px] top-1 w-3 h-3 rounded-full bg-primary-500 ring-4 ring-primary-500/20"></span> <div class="mb-1.5"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2020 – 2023</span> <span class="ml-2 text-base font-bold text-gray-900 dark:text-white">规模化生产处理 Production-scale Processing</span> </div> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">批量视频处理 · Batch Processing</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">GPU 加速 · GPU Acceleration</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">质量控制 · Quality Control</span> </div> </div> <div class="relative"> <span class="absolute -left-[31px] top-1 w-3 h-3 rounded-full bg-primary-500 ring-4 ring-primary-500/20"></span> <div class="mb-1.5"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2023 – 2024</span> <span class="ml-2 text-base font-bold text-gray-900 dark:text-white">Transformer 修复 Transformer-based Restoration</span> </div> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">Transformer</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">U-Net</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">参考引导 · Reference Guidance</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">4K 增强 · 4K Enhancement</span> </div> </div> <div class="relative"> <span class="absolute -left-[31px] top-1 w-3 h-3 rounded-full bg-primary-500 ring-4 ring-primary-500/20"></span> <div class="mb-1.5"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2024 – 2025</span> <span class="ml-2 text-base font-bold text-gray-900 dark:text-white">端到端智能修复 End-to-End Restoration</span> </div> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">语义先验 · Semantic Prior</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">感知损失 · Perceptual Loss</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">时序一致性 · Temporal Consistency</span> </div> </div> </div> <h2 class="vr-sec"><span class="vr-sec-cn">核心能力</span><span class="vr-sec-en">Core Capabilities</span></h2> <div class="grid gap-4 sm:grid-cols-2 lg:grid-cols-3"> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <h3 class="text-base font-bold text-gray-900 dark:text-white mb-3">图像复原 Image Restoration</h3> <ul class="space-y-1.5"> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>图像去噪 · Image Denoising</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>缺陷修复 · Defect Removal</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>去模糊 · Deblurring</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>画质增强 · Image Enhancement</span> </li> </ul> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <h3 class="text-base font-bold text-gray-900 dark:text-white mb-3">视频修复 Video Restoration</h3> <ul class="space-y-1.5"> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>时序一致性 · Temporal Consistency</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>视频去噪 · Video Denoising</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>逐帧修复 · Frame-level Restoration</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>长视频处理 · Long-form Video Processing</span> </li> </ul> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <h3 class="text-base font-bold text-gray-900 dark:text-white mb-3">超分辨率 Super Resolution</h3> <ul class="space-y-1.5"> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>分辨率增强 · Resolution Enhancement</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>4K 修复 · 4K Restoration</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>参考引导重建 · Reference-guided Reconstruction</span> </li> </ul> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <h3 class="text-base font-bold text-gray-900 dark:text-white mb-3">深度学习 Deep Learning</h3> <ul class="space-y-1.5"> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>CNN</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>U-Net</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>Transformer</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>感知优化 · Perceptual Optimization</span> </li> </ul> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <h3 class="text-base font-bold text-gray-900 dark:text-white mb-3">数据集工程 Dataset Engineering</h3> <ul class="space-y-1.5"> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>退化模拟 · Degradation Simulation</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>训练集构建 · Training Dataset</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>评估集 · Evaluation Dataset</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>质量评估 · Quality Assessment</span> </li> </ul> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <h3 class="text-base font-bold text-gray-900 dark:text-white mb-3">生产工程 Production Engineering</h3> <ul class="space-y-1.5"> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>GPU 处理 · GPU Processing</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>批量调度 · Batch Scheduling</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>大规模媒体处理 · Large-scale Processing</span> </li> <li class="flex items-start gap-2 text-sm text-gray-600 dark:text-gray-400"> <span class="mt-1.5 w-1.5 h-1.5 rounded-full bg-primary-500 shrink-0"></span> <span>质量控制 · Quality Control</span> </li> </ul> </div> </div> <h2 class="vr-sec"><span class="vr-sec-cn">代表性项目</span><span class="vr-sec-en">Selected Production Projects</span></h2> <div class="grid gap-4 sm:grid-cols-2 lg:grid-cols-2"> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <div class="mb-2"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2015.04 – 2015.08</span> <h4 class="mt-1 text-base font-bold text-gray-900 dark:text-white">《本命年》电影画质升级翻新</h4> </div> <p class="text-sm text-gray-600 dark:text-gray-400 mb-3 leading-relaxed">针对老电影胶片素材中的噪声、模糊与色彩退化问题开展画质升级与翻新处理，负责图像降噪、画质增强算法调试、素材预处理及效果迭代，完成翻新成片输出。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">图像复原</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">视频去噪</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">胶片修复</span> </div> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <div class="mb-2"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2015 – 2021</span> <h4 class="mt-1 text-base font-bold text-gray-900 dark:text-white">中影数字基地 / 北京电影资料馆馆藏老电影修复</h4> </div> <p class="text-sm text-gray-600 dark:text-gray-400 mb-3 leading-relaxed">面向馆藏胶片老电影开展数字化修复，处理霉斑、划痕、颗粒噪声、帧闪烁、画面抖动等典型胶片损伤；负责图像修复与视频增强算法研发调优，并参与数据集构建、效果评估和批量处理质量控制。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">CNN</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">图像复原</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">时序处理</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">批量处理</span> </div> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <div class="mb-2"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2017</span> <h4 class="mt-1 text-base font-bold text-gray-900 dark:text-white">《军师联盟》剧集画质增强</h4> </div> <p class="text-sm text-gray-600 dark:text-gray-400 mb-3 leading-relaxed">针对剧集素材噪声特征开展视频降噪和画质增强算法适配，进行空间与时序联合处理及效果迭代。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">视频增强</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">时序处理</span> </div> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <div class="mb-2"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2017 – 2018</span> <h4 class="mt-1 text-base font-bold text-gray-900 dark:text-white">《厉害了，我的国》历史镜头修复</h4> </div> <p class="text-sm text-gray-600 dark:text-gray-400 mb-3 leading-relaxed">针对影片中的历史档案镜头进行画质修复和清晰度增强，处理模糊、噪声及色彩偏差等问题，并完成多轮效果优化。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">超分辨率</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">图像复原</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">历史影像</span> </div> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <div class="mb-2"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2020 – 2022</span> <h4 class="mt-1 text-base font-bold text-gray-900 dark:text-white">广东广播电视台剧集与纪录片修复</h4> </div> <p class="text-sm text-gray-600 dark:text-gray-400 mb-3 leading-relaxed">面向老剧集和历史纪录片开展批量画质修复与增强，针对老素材噪声和色彩失真进行算法适配、参数调优和批量输出质量控制。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">视频修复</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">色彩校正</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">GPU 处理</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">批量处理</span> </div> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <div class="mb-2"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2021</span> <h4 class="mt-1 text-base font-bold text-gray-900 dark:text-white">华数数字电视传媒纪录片修复</h4> </div> <p class="text-sm text-gray-600 dark:text-gray-400 mb-3 leading-relaxed">针对不同来源纪录片素材的退化特征进行视频复原、去噪与色彩还原。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">视频修复</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">去噪</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">色彩还原</span> </div> </div> <div class="bg-white dark:bg-white/[0.03] border border-gray-200 dark:border-white/10 rounded-2xl p-5 h-full"> <div class="mb-2"> <span class="font-mono text-xs font-semibold text-primary-600 dark:text-primary-400">2023</span> <h4 class="mt-1 text-base font-bold text-gray-900 dark:text-white">海南影视 / 琼剧历史素材修复</h4> </div> <p class="text-sm text-gray-600 dark:text-gray-400 mb-3 leading-relaxed">针对老录像带、历史纪录片及传统戏曲影像进行视频降噪、色彩校正和超分辨率增强。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">视频修复</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">色彩校正</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">超分辨率</span> </div> </div> </div> <h2 class="vr-sec"><span class="vr-sec-cn">研究与创新</span><span class="vr-sec-en">Research & Innovation</span></h2> <div class="vr-research"> <h3 id="时光裂痕智能修复唤醒沉睡的记忆">时光裂痕：智能修复唤醒沉睡的记忆</h3> <p><strong>2024 – 2025 · AI-powered Video Restoration</strong></p> <p>面向老旧影视资料研发端到端智能修复计算平台，探索从缺陷修复到超分辨率增强的一体化视频复原方案。</p> <div class="highlight"><pre tabindex="0" class="chroma"><code class="language-fallback" data-lang="fallback"><span class="line"><span class="cl">Cascade Deep Neural Networks </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Restoration + Super Resolution </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Perceptual Loss </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Semantic Prior </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Temporal Illumination Consistency </span></span><span class="line"><span class="cl"> ↓ </span></span><span class="line"><span class="cl">Video Restoration </span></span></code></pre></div><p>参与级联深度神经网络模块调优、修复与超分模块级联调试、帧间光照一致性约束优化，并参与高层语义先验、数据集构建、模型验证和算法模块落地。</p> <p><strong>个人角色：核心算法人员 · 项目答辩主讲人</strong></p> <div class="rounded-2xl border-2 border-amber-400 dark:border-amber-500/50 bg-amber-50 dark:bg-amber-500/10 p-6 h-full"> <div class="flex items-center gap-2"> <span class="text-2xl leading-none">🏆</span> <span class="text-2xl font-extrabold text-amber-700 dark:text-amber-300">二等奖</span> </div> <div class="mt-2 text-sm font-semibold text-amber-900 dark:text-amber-200">智能修复与增强方案赛道</div> <div class="mt-1 text-xs text-gray-600 dark:text-gray-400 leading-relaxed">MediaAIAC · 第四届广播电视和网络视听人工智能应用创新大赛</div> <div class="mt-1 font-mono text-xs text-gray-500 dark:text-gray-500">2025.09</div> </div> </div> <div class="vr-research vr-research-sub"> <h3 id="影视媒体修复与视觉质量提升">影视媒体修复与视觉质量提升</h3> <p><em>基于级联深度神经网络和高层语义信息</em></p> <p>参与基于级联深度神经网络和高层语义信息的影视媒体修复与视觉质量提升项目，围绕影视媒体缺陷修复、视觉质量增强等方向参与算法研发与项目实施。</p> <p><strong>个人角色：项目参与人</strong></p> <div class="rounded-2xl border border-gray-200 dark:border-white/10 bg-white dark:bg-white/[0.03] p-5 h-full"> <div class="flex items-center gap-2"> <span class="text-lg leading-none">🏆</span> <span class="text-lg font-bold text-gray-800 dark:text-gray-200">三等奖</span> </div> <div class="mt-1.5 text-sm text-gray-600 dark:text-gray-400">视频修复类</div> <div class="mt-1 text-xs text-gray-500 dark:text-gray-500 leading-relaxed">MediaAIAC · 首届广播电视和网络视听人工智能应用创新大赛</div> <div class="mt-1 font-mono text-xs text-gray-400 dark:text-gray-500">2021.05</div> </div> </div> <h2 class="vr-sec"><span class="vr-sec-cn">长期技术实践</span><span class="vr-sec-en">Long-term Practice</span></h2> <div class="vr-practice"> <p>长期参与历史影视、老照片及音视频资料的智能修复与画质增强，持续迭代 CNN、Transformer 等图像复原模型，并围绕模拟退化数据、模型训练、效果评估、GPU 处理和质量控制建立可复用的算法与工程流程。</p> <div class="flex flex-wrap gap-1.5"> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">CNN</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">Transformer</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">图像复原 · Image Restoration</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">超分辨率 · Super Resolution</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">视频处理 · Video Processing</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">数据集工程 · Dataset Engineering</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">GPU 计算 · GPU Computing</span> <span class="px-2.5 py-1 text-xs font-medium rounded-full bg-primary-50 dark:bg-primary-500/10 text-primary-700 dark:text-primary-300 border border-primary-200 dark:border-primary-500/20">质量评估 · Quality Assessment</span> </div> </div> </div>