<?xml version="1.0" encoding="utf-8" ?><rss version="2.0"><channel><title><![CDATA[JustWantToLearn的博客]]></title><description><![CDATA[]]></description><link>https://blog.youkuaiyun.com/JustWantToLearn</link><language>zh-cn</language><generator>https://blog.youkuaiyun.com/</generator><copyright><![CDATA[Copyright &copy; JustWantToLearn]]></copyright><item><title><![CDATA[MOFA-Video: Controllable Image Animation via Generative Motion Field Adaptions in Frozen Image-to-Vi]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/141125208</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/141125208</guid><author>JustWantToLearn</author><pubDate>Sun, 13 Oct 2024 11:02:57 +0800</pubDate><description><![CDATA[MOFA-Video，一种先进的可控图像利用各种附加可控信号(如人类地标参考、手动轨迹，另一个甚至提供视频)或它们的组合。这不同于以往的方法只能在视频生成流水线中工作。对于MOFA-Adapters，在给定的稀疏控制条件下，首先对视频的时间运动一致性进行分析，生成密集的运动流，将给定图像的多尺度特征作为引导特征进行包裹，以稳定地生成视频扩散。我们为手动轨迹和人类地标分别训练两个运动适配器，因为它们都包含关于控制的稀疏信息。经过培训，不同领域的MOFA-Adapters也可以共同努力，实现更可控的视频生成。]]></description><category></category></item><item><title><![CDATA[快手LivePortrait: Efficient Portrait Animation with Stitching and Retargeting Control]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/140833411</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/140833411</guid><author>JustWantToLearn</author><pubDate>Sun, 13 Oct 2024 11:01:18 +0800</pubDate><description><![CDATA[快手LivePortrait: Efficient Portrait Animation with Stitching and Retargeting Control]]></description><category></category></item><item><title><![CDATA[视频合成—NeurIPS2018—Video-to-Video Synthesis]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139247456</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139247456</guid><author>JustWantToLearn</author><pubDate>Sun, 09 Jun 2024 11:49:46 +0800</pubDate><description><![CDATA[在本文中，提出了一种在生成对抗学习框架下的视频到视频合成方法。通过精心设计的生成器和鉴别器，再加上时空对抗目标，我们在包括分割蒙版、草图和姿势在内的多种输入格式上实现了高分辨率、逼真、时间连贯的视频结果。]]></description><category></category></item><item><title><![CDATA[换脸的双重防御—2024TIFS—Dual Defense: Adversarial, Traceable, and Invisible Robust Watermarking Against Fac]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139550697</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139550697</guid><author>JustWantToLearn</author><pubDate>Sat, 08 Jun 2024 21:37:09 +0800</pubDate><description><![CDATA[一种基于鲁棒对抗性水印的创新框架。它同时跟踪图像版权，并通过一次性嵌入鲁棒对抗水印来破坏人脸交换模型。]]></description><category></category></item><item><title><![CDATA[面试真题：语音特征提取方式——为何要这样提取？]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139222654</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139222654</guid><author>JustWantToLearn</author><pubDate>Sun, 26 May 2024 23:24:35 +0800</pubDate><description><![CDATA[语音特征提取的方式，为什么要这样提取]]></description><category></category></item><item><title><![CDATA[腾讯音乐TME-MuseTalk模型代码详解]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139070253</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139070253</guid><author>JustWantToLearn</author><pubDate>Sun, 26 May 2024 19:57:12 +0800</pubDate><description><![CDATA[MuseTalk的测试流程]]></description><category></category></item><item><title><![CDATA[代码解析—part3 训练ADM—CVPR2023—Implicit Identity Leakage: The Stumbling Block to Improving Deepfake]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139116455</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139116455</guid><author>JustWantToLearn</author><pubDate>Thu, 23 May 2024 15:16:07 +0800</pubDate><description><![CDATA[论文代码训练模块]]></description><category></category></item><item><title><![CDATA[代码解析—part 2 数据集加载MFS—CVPR2023—Implicit Identity Leakage: The Stumbling Block to Improving Deepfake]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139092687</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139092687</guid><author>JustWantToLearn</author><pubDate>Wed, 22 May 2024 18:44:13 +0800</pubDate><description><![CDATA[MFS模块代码解析]]></description><category></category></item><item><title><![CDATA[面试真题——self-attention 不要只记公式了！！]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138874373</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138874373</guid><author>JustWantToLearn</author><pubDate>Sun, 19 May 2024 15:10:42 +0800</pubDate><description><![CDATA[面试真题-手撕self-attention]]></description><category></category></item><item><title><![CDATA[AIGC这是我想进入就能进的吗——研二找暑期实习AIGC方向严重受挫]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139013748</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/139013748</guid><author>JustWantToLearn</author><pubDate>Fri, 17 May 2024 22:03:49 +0800</pubDate><description><![CDATA[AIGC说的我们想进去，就真能收留我们一样，你有硬件训练吗，你有数据吗？谨慎考虑！！！别入坑了爬不出来了]]></description><category></category></item><item><title><![CDATA[代码解析—part1 数据准备—CVPR2023—Implicit Identity Leakage: The Stumbling Block to Improving Deepfake]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138773005</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138773005</guid><author>JustWantToLearn</author><pubDate>Mon, 13 May 2024 00:05:13 +0800</pubDate><description><![CDATA[代码理解]]></description><category></category></item><item><title><![CDATA[deepfake检测——Implicit Identity Leakage: The Stumbling Block to Improving Deepfake Detection Ge]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138758033</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138758033</guid><author>JustWantToLearn</author><pubDate>Mon, 13 May 2024 00:02:58 +0800</pubDate><description><![CDATA[在本文中，分析了二元分类器在深度伪造检测任务中的泛化能力。发现阻碍它们泛化的绊脚石是由意想不到的图像学习身份表示引起的。这种现象被称为隐式身份泄漏，基于这样的理解，我们提出了一种简单而有效的方法，以减少这种现象的影响。大量的实验结果表明，我们的方法在数据集内和跨数据集评估方面都优于最先进的方法。]]></description><category></category></item><item><title><![CDATA[diffusion model基础学习——DDPM代码解析]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138253376</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138253376</guid><author>JustWantToLearn</author><pubDate>Sat, 11 May 2024 10:28:56 +0800</pubDate><description><![CDATA[本文主要梳理以训练CIFAR-10数据集【包含10个类别的60000个32x32彩色图像】为例，DDPM与DDPM+condition的pytorch代码训练流程，模型搭建过程。细节方面解释会在代码中标注释，建议先自行查看一下代码流程。]]></description><category></category></item><item><title><![CDATA[AIGC生成论文汇总]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138683666</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138683666</guid><author>JustWantToLearn</author><pubDate>Fri, 10 May 2024 22:01:21 +0800</pubDate><description><![CDATA[AIGC生成论文汇总]]></description><category></category></item><item><title><![CDATA[面试收获——对于Batch Normalization不要只会公式了！！！！]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138681798</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138681798</guid><author>JustWantToLearn</author><pubDate>Fri, 10 May 2024 21:33:19 +0800</pubDate><description><![CDATA[batch normalization 深度理解]]></description><category></category></item><item><title><![CDATA[AIGC论文阅读——[Visual Speech-Aware Perceptual 3D Facial Expression Reconstruction from Videos]]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138605835</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/138605835</guid><author>JustWantToLearn</author><pubDate>Thu, 09 May 2024 16:25:28 +0800</pubDate><description><![CDATA[提出了第一种3D说话头的视觉语音感知重建方法。该方法不依赖于文本转录或音频;相反，采用了一种“唇读”损失，它指导训练过程，以增加对嘴的感知。广泛的主观和客观评估已经证实，3D重建的结果明显优于仅依赖于嘴部运动的几何损失的对应方法，以及使用直接3D监督的方法。]]></description><category></category></item><item><title><![CDATA[WTAGRAPH: Web Tracking and Advertising Detection using Graph Neural NetworksTOC]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/130861904</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/130861904</guid><author>JustWantToLearn</author><pubDate>Tue, 30 May 2023 21:09:40 +0800</pubDate><description><![CDATA[如今，网络跟踪和广告 (WTA) 在网络上无处不在，不断损害用户的隐私。现有的防御解决方案，例如广泛部署的基于过滤器列表的阻止工具和先前研究中提出的基于替代机器学习的解决方案，在准确性和有效性方面存在局限性。在这项工作中，我们提出了 WTAGRAPH，**一种基于图神经网络 (GNN) 的网络跟踪和广告检测框架**。我们首先构造一个表示 HTTP 网络流量的属性同质多图 (AHMG)，并将网络跟踪和广告检测制定为 AHMG 中基于 GNN 的边缘表示学习和分类的任务]]></description><category></category></item><item><title><![CDATA[KMP算法]]></title><link>https://blog.youkuaiyun.com/JustWantToLearn/article/details/103011815</link><guid>https://blog.youkuaiyun.com/JustWantToLearn/article/details/103011815</guid><author>JustWantToLearn</author><pubDate>Mon, 11 Nov 2019 16:21:48 +0800</pubDate><description><![CDATA[KMP算法
设有两个串s和t，在串s中找到一个与t相等的子串。通常把s称为目标串，把t称为模式串。在目标串s中找到了一个模式串t则模式串匹配成功，没有找到则匹配不成功。
设n为串s的长度，m为串t的长度。
为什么要使用KMP算法？
模式匹配串是一个比较复杂的串操作，用暴力的方法来做的话，复杂度为O（m*n）,但如果使用KMP算法，复杂度为O（m+n），大大提高了算法效率。
从模式串t中提取出加速匹...]]></description><category></category></item></channel></rss>