https://www.ruyaohui.com/Recent content on RiverYu's Homehttps://www.ruyaohui.com/images/cover.pnghttps://www.ruyaohui.com/images/cover.pngHugo -- 0.152.2zh-cn2026 RiverYu&rsquo;s Home · Per aspera ad astraThu, 18 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/styleganv3/Thu, 18 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/styleganv3/傅里叶特征https://www.ruyaohui.com/posts/latent-space-tokenizer-to-aigc-detection/Tue, 16 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/latent-space-tokenizer-to-aigc-detection/梳理 RAE、Causal 3D VAE 及 Cosmos Tokenizerhttps://www.ruyaohui.com/posts/bim/Mon, 15 Jun 2026 00:00:00 +0000https://www.ruyaohui.com/posts/bim/<h2 id="1-问题背景慢放视频会导致卡顿">1. 问题背景：慢放视频会导致卡顿</h2> <p>视频由连续图像帧组成，帧率表示每秒播放的画面数量，单位是 fps。帧率越高，画面变化越连续，视觉观感通常越流畅。</p> <p>当一段视频被慢放时，如果只是简单拉长播放时间，而没有增加新的中间帧，视频的有效帧率会下降。例如一段 30 fps 的视频被放慢到 0.25 倍，实际观感可能接近 7.5 fps，画面就会出现明显卡顿。</p>https://www.ruyaohui.com/posts/chordedit-optimal-transport-image-editing/Thu, 11 Jun 2026 22:00:00 +0800https://www.ruyaohui.com/posts/chordedit-optimal-transport-image-editing/为什么一步图像编辑会因为高能量漂移场而失稳，如何用动态最优传输、局部二次估计和时间平滑构造稳定的 Chord Control Fieldhttps://www.ruyaohui.com/posts/deep-learning-optimizers-sgd-to-adamw/Tue, 09 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/deep-learning-optimizers-sgd-to-adamw/SGD、Momentum、RMSProp 到 AdamWhttps://www.ruyaohui.com/posts/network-troubleshooting-from-lag-to-layer/Mon, 08 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/network-troubleshooting-from-lag-to-layer/从 DNS、TCP、TLS 到 HTTP、TUN、MTU 与浏览器的分层网络排障方法https://www.ruyaohui.com/posts/fgsm/Sun, 07 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/fgsm/对抗笔记https://www.ruyaohui.com/posts/math/Sat, 06 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/math/数学整理https://www.ruyaohui.com/posts/encoder-deep-dive-aigc-video-detection/Wed, 03 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/posts/encoder-deep-dive-aigc-video-detection/Encoder 如何从像素层抽象出证据空间，并结合 GRU 与 Normalization 构建鲁棒的视频检测框架。https://www.ruyaohui.com/notes/os_xyx/Mon, 01 Jun 2026 10:00:00 +0800https://www.ruyaohui.com/notes/os_xyx/从状态机的视角出发，全面梳理并发、虚拟化与持久化三大核心机制，并记录 OS 实验中的踩坑与顿悟。https://www.ruyaohui.com/posts/transformer-stability-to-aigc-video-detection/Thu, 28 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/transformer-stability-to-aigc-video-detection/跨界迁移到 AIGC 视频检测的课题中https://www.ruyaohui.com/posts/systematic-research-nomad-to-builder/Wed, 27 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/systematic-research-nomad-to-builder/从系统性科研、学术标签到大兵团协作，本文是对上海交通大学杨学老师关于“科研长期主义”访谈的深度思考。https://www.ruyaohui.com/posts/ai-model-reproduction-engineering-tips/Tue, 26 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/ai-model-reproduction-engineering-tips/总结了自然排序、占位数据、可回滚补丁等技巧。https://www.ruyaohui.com/posts/brain-power-upgrade-research-system/Sun, 24 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/brain-power-upgrade-research-system/建立一套从供能到输出的长期系统。https://www.ruyaohui.com/posts/randy-pausch-last-lecture-and-research-path/Thu, 21 May 2026 18:00:00 +0800https://www.ruyaohui.com/posts/randy-pausch-last-lecture-and-research-path/真正的强大，是把聪明变成创造路径的能力，把梦想变成具体行动，把别人给的帮助变成继续帮助更多人的责任。从《最后的演讲》中寻找科研与人生的答案。https://www.ruyaohui.com/posts/math-modeling-gdp-employment-optimization/Wed, 20 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/math-modeling-gdp-employment-optimization/把现实问题拆解为目标函数与约束条件。复盘如何一步步将工业投资分配问题，打磨成一个混合整数规划模型，并反思最终交付的教训。https://www.ruyaohui.com/posts/aigc-dectector/Mon, 18 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/aigc-dectector/关于方向的思考和整理https://www.ruyaohui.com/posts/docx-omml-formula-automation/Sat, 16 May 2026 22:00:00 +0800https://www.ruyaohui.com/posts/docx-omml-formula-automation/通过 Python 直接操作 .docx 底层结构，将文本批量转换为 Word 原生公式。https://www.ruyaohui.com/posts/shunyu-yao-interview-reflection/Fri, 15 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/shunyu-yao-interview-reflection/学习到如何快速成长https://www.ruyaohui.com/posts/mathplay/Tue, 12 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/mathplay/想想该怎么做数学建模https://www.ruyaohui.com/notes/tmux-under-the-hood-os-principles/Sun, 10 May 2026 22:00:00 +0800https://www.ruyaohui.com/notes/tmux-under-the-hood-os-principles/从 PTY 伪终端、SIGHUP 信号到进程树模型https://www.ruyaohui.com/posts/3d-vision-dependency-hell-refactoring/Sun, 10 May 2026 15:00:00 +0800https://www.ruyaohui.com/posts/3d-vision-dependency-hell-refactoring/标准环境搭建实践。https://www.ruyaohui.com/posts/ai-gold-rush-and-research-automation/Sun, 10 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/ai-gold-rush-and-research-automation/从 AI 淘金热到底层工具链，关于过去与未来https://www.ruyaohui.com/posts/5g-misunderstood-infrastructure-ai-era/Fri, 08 May 2026 22:00:00 +0800https://www.ruyaohui.com/posts/5g-misunderstood-infrastructure-ai-era/5G与AIhttps://www.ruyaohui.com/notes/first-principles-3d-virtual-human/Fri, 08 May 2026 10:00:00 +0800https://www.ruyaohui.com/notes/first-principles-3d-virtual-human/从第一性原理出发，系统性复盘 3D 虚拟人如何在硅基世界中“重塑肉身”并“注入灵魂”，《基于神经辐射场与高斯溅射的3D虚拟人重建与驱动方法综述》https://www.ruyaohui.com/posts/world-cup-2026-broadcasting-rights-dispute/Thu, 07 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/world-cup-2026-broadcasting-rights-dispute/写在 2026 世界杯转播权僵局之时。https://www.ruyaohui.com/posts/reading-note-hongwa-heiwa/Wed, 06 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/reading-note-hongwa-heiwa/它写出了一种带有时代感的青春，缓慢、复杂、充满挣扎。只是它的结尾，似乎没有完全承接住前面积累起来的重量。https://www.ruyaohui.com/posts/podcast-reflection-ai-paradigm-shift/Mon, 04 May 2026 10:00:00 +0800https://www.ruyaohui.com/posts/podcast-reflection-ai-paradigm-shift/在 AI 范式剧烈变化的时候，一个人到底该怎么做事？https://www.ruyaohui.com/posts/tencent-ai-competition-youying-blueprint/Thu, 30 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/tencent-ai-competition-youying-blueprint/面对腾讯创新赛，我构思了一款名为【有应】的视觉社交产品。AI 不应替代人类陪伴，而应帮我们找到愿意回应的人，为“社交落空”提供终极兜底。https://www.ruyaohui.com/posts/rebuilding-personal-os-generative-mindset/Wed, 29 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/rebuilding-personal-os-generative-mindset/大多数人的努力只是在别人设定的闭环系统里“刷分”。停止在开放世界里做判别式的多选题，用极客思维重构你的底层 Loss Function。https://www.ruyaohui.com/notes/how-to-read-cv-math-formulas/Mon, 27 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/notes/how-to-read-cv-math-formulas/学习数学公式https://www.ruyaohui.com/notes/resnet-unet-resunet-explained/Thu, 23 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/notes/resnet-unet-resunet-explained/为什么 U-Net 和 ResNet 看起来很像？Skip Connection 在解决什么问题？https://www.ruyaohui.com/notes/agaigc/Sun, 19 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/notes/agaigc/ICCV 2023 Workshop 论文精读https://www.ruyaohui.com/posts/research-note-gan-aigc-evasion/Fri, 17 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/research-note-gan-aigc-evasion/GAN-RESTARThttps://www.ruyaohui.com/posts/ec/Tue, 14 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/ec/记录主英辅中”双语版本https://www.ruyaohui.com/posts/flashavatar-setup/Sun, 12 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/flashavatar-setup/放弃Conda，使用Micromambahttps://www.ruyaohui.com/posts/computer-science-career-outlook-2028-2032/Sat, 11 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/computer-science-career-outlook-2028-2032/热爱最重要，让自己成为不可替代的人https://www.ruyaohui.com/posts/os-m3-sperf-reflection/Fri, 10 Apr 2026 15:00:00 +0800https://www.ruyaohui.com/posts/os-m3-sperf-reflection/如何用 pipe、fork 和 execve 串起一条优雅的分析流水线。https://www.ruyaohui.com/posts/research-log-3-dynamic-modeling-pivot/Thu, 09 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/research-log-3-dynamic-modeling-pivot/导师的指点https://www.ruyaohui.com/posts/king-of-comedy-review/Wed, 08 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/king-of-comedy-review/其实，我是个演员https://www.ruyaohui.com/posts/3dgs-ex/Tue, 07 Apr 2026 12:00:00 +0800https://www.ruyaohui.com/posts/3dgs-ex/这篇文章记录的不是一份完美成果，而是一段比较真实的工程化科研推进过程：从官方 baseline 验证，到自有视频数据处理，再到输入契约、相机模型、COLMAP 主模型选择，以及最终主线数据集的收敛。https://www.ruyaohui.com/posts/research-log-2-experiment-records/Mon, 06 Apr 2026 18:00:00 +0800https://www.ruyaohui.com/posts/research-log-2-experiment-records/从验证官方 Baseline 到发现数据链路中的系统性剥离问题。https://www.ruyaohui.com/posts/music-sharing-experiment/Sun, 05 Apr 2026 12:00:00 +0800https://www.ruyaohui.com/posts/music-sharing-experiment/朋友圈里的音乐分享，首先是一张社交名片，其次才是一段内容。读者可以自行尝试30天，https://www.ruyaohui.com/posts/3dgs-pipeline/Sat, 04 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/3dgs-pipeline/从泛泛的想法到真实的实验https://www.ruyaohui.com/notes/gphm-paper-reading/Thu, 02 Apr 2026 12:00:00 +0800https://www.ruyaohui.com/notes/gphm-paper-reading/GPHM 论文阅读https://www.ruyaohui.com/posts/monocular-full-head-reconstruction/Wed, 01 Apr 2026 10:00:00 +0800https://www.ruyaohui.com/posts/monocular-full-head-reconstruction/一些感悟https://www.ruyaohui.com/posts/monocular/Tue, 31 Mar 2026 10:00:00 +0800https://www.ruyaohui.com/posts/monocular/行胜于言https://www.ruyaohui.com/posts/2025-review/Fri, 02 Jan 2026 14:00:00 +0800https://www.ruyaohui.com/posts/2025-review/身体的健康下降了很多，人际关系也没有处理好，但车到山前必有路。回顾这混沌与挣扎并存的一年。https://www.ruyaohui.com/posts/2026-goals/Thu, 01 Jan 2026 10:00:00 +0800https://www.ruyaohui.com/posts/2026-goals/这一年，我要充分发挥我的主观能动性，不让自己后悔。14 个 Flag，立帖为证。https://www.ruyaohui.com/notes/code-comparison-papera-vs-paperb/Thu, 27 Nov 2025 19:40:47 +0800https://www.ruyaohui.com/notes/code-comparison-papera-vs-paperb/梦开始的地方：A deep dive comparison between two classic CNN architectures.https://www.ruyaohui.com/projects/kaggle-digit-recognizer/Sun, 23 Nov 2025 11:20:00 +0800https://www.ruyaohui.com/projects/kaggle-digit-recognizer/基于 CNN (LeNet-5) 实现的手写数字识别项目。 完成了数据预处理、模型搭建与训练，最终在 Kaggle 取得 Top 15% 的成绩。 包含完整的 PyTorch 代码复现与参数调优记录。https://www.ruyaohui.com/about/Mon, 01 Jan 0001 00:00:00 +0000https://www.ruyaohui.com/about/<h2 id="hi-im-riveryu">Hi, I&rsquo;m RiverYu</h2> <p>I am an undergraduate student majoring in <strong>CS (Network Engineering)</strong> at <strong>Henan Normal University</strong>.<br> My research interests lie in <strong>Computer Vision</strong>, the way machines see the world, and perhaps one way we imagine the future of computing.</p> <h3 id="research-experience">Research Experience</h3> <ul> <li><strong>Undergraduate Researcher</strong> in the <strong>3D Vision Group</strong>.</li> <li>My research focuses on <strong>3D Vision</strong> and <strong>Representation Learning</strong>, especially <strong>3D Generation &amp; Editing</strong>, <strong>3D Reconstruction</strong>,and <strong>Latent Representations in Generative Models</strong>.</li> </ul> <h3 id="goals">Goals</h3> <ul> <li>Build solid foundations in <strong>Mathematics</strong> and <strong>Computer Science</strong></li> <li>Complete high-quality projects for research and real-world applications</li> <li>Prepare for future graduate study in Computer Science</li> </ul> <h3 id="tech-stack">Tech Stack</h3> <ul> <li><strong>Languages:</strong> Python, C++</li> <li><strong>Frameworks:</strong> PyTorch, CUDA</li> <li><strong>Tools:</strong> Git, Linux, Hugo, LaTeX</li> </ul>https://www.ruyaohui.com/search/Mon, 01 Jan 0001 00:00:00 +0000https://www.ruyaohui.com/search/