https://111qqz.com/en/tags/diffusion/Recent content in Diffusion on 111qqz's blogHugo -- gohugo.ioenhust.111qqz@gmail.com (111qqz)hust.111qqz@gmail.com (111qqz)© 2011-2026 111qqzTue, 13 Oct 2026 20:00:00 +0800https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-10-13-ddpm-reverse-denoising-learning/Tue, 13 Oct 2026 20:00:00 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-10-13-ddpm-reverse-denoising-learning/<h2 class="relative group">接着上一篇的问题 <div id="接着上一篇的问题" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e6%8e%a5%e7%9d%80%e4%b8%8a%e4%b8%80%e7%af%87%e7%9a%84%e9%97%ae%e9%a2%98" aria-label="Anchor">#</a> </span> </h2> <p><a href="https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-10-13-ddpm-forward-gaussian-skip/" >上一篇</a>搞清楚了一件事：DDPM 的前向扩散虽然定义成一步一步加噪的马尔可夫链，但因为高斯分布在线性组合下的封闭性，任意时刻的 \(x_t\) 都可以一步算出来：</p>https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-10-13-ddpm-forward-gaussian-skip/Tue, 13 Oct 2026 19:20:00 +0800hust.111qqz@gmail.com (111qqz)https://111qqz.com/en/post/%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0/2026-10-13-ddpm-forward-gaussian-skip/<h2 class="relative group">起因 <div id="起因" class="anchor"></div> <span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100 select-none"> <a class="text-primary-300 dark:text-neutral-700 !no-underline" href="#%e8%b5%b7%e5%9b%a0" aria-label="Anchor">#</a> </span> </h2> <p>最近在看 DDPM（Denoising Diffusion Probabilistic Models）。前向扩散的定义写得很清楚：从原始图片出发，每一步往上面加一点噪声，一步一步把图片变成纯噪声。</p>