Buckets:

hf-doc-build
/

doc-dev

hf-doc-build/doc-dev / diffusers /pr_13881 /zh /_app /immutable /nodes /5.sScpQL_r.js

HuggingFaceDocBuilder's picture

HuggingFaceDocBuilder

23 days ago

59.3 kB

	import"../chunks/DsnmJJEf.js";import{i as v,h as S,C as z,H as a,a as s,E as x,s as H}from"../chunks/DdZvggmf.js";import{p as D,o as q,s as l,f as P,a as F,b as L,c as Y,n as K}from"../chunks/BbekZcyp.js";const O='{"title":"Diffusion模型评估指南","local":"diffusion模型评估指南","sections":[{"title":"评估场景","local":"评估场景","sections":[],"depth":2},{"title":"定性评估","local":"定性评估","sections":[],"depth":2},{"title":"定量评估","local":"定量评估","sections":[{"title":"文本引导图像生成","local":"文本引导图像生成","sections":[],"depth":3},{"title":"图像条件式文本生成图像","local":"图像条件式文本生成图像","sections":[],"depth":3},{"title":"基于类别的图像生成","local":"基于类别的图像生成","sections":[],"depth":3}],"depth":2}],"depth":1}';var $=Y('<meta name="hf:doc:metadata"/>'),ll=Y(`<p></p> <!> <!> <a target="_blank" href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/evaluation.ipynb"><img src="https://colab.research.google.com/assets/colab-badge.svg" alt="在 Colab 中打开"/></a> <blockquote class="tip"><p>鉴于当前已出现针对图像生成Diffusion模型的成熟评估框架（如<a href="https://crfm.stanford.edu/helm/heim/latest/" rel="nofollow">HEIM</a>、<a href="https://huggingface.co/papers/2307.06350" rel="nofollow">T2I-Compbench</a>、<a href="https://huggingface.co/papers/2310.11513" rel="nofollow">GenEval</a>），本文档部分内容已过时。</p></blockquote> <p>像 <a href="https://huggingface.co/docs/diffusers/stable_diffusion" rel="nofollow">Stable Diffusion</a> 这类生成模型的评估本质上是主观的。但作为开发者和研究者，我们经常需要在众多可能性中做出审慎选择。那么当面对不同生成模型（如 GANs、Diffusion 等）时，该如何决策？</p> <p>定性评估容易产生偏差，可能导致错误结论；而定量指标又未必能准确反映图像质量。因此，通常需要结合定性与定量评估来获得更可靠的模型选择依据。</p> <p>本文档将系统介绍扩散模型的定性与定量评估方法（非穷尽列举）。对于定量方法，我们将重点演示如何结合 <code>diffusers</code> 库实现这些评估。</p> <p>文档所示方法同样适用于评估不同<a href="https://huggingface.co/docs/diffusers/main/en/api/schedulers/overview" rel="nofollow">噪声调度器</a>在固定生成模型下的表现差异。</p> <!> <p>我们涵盖以下Diffusion模型管线的评估：</p> <ul><li>文本引导图像生成（如 <a href="https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/text2img" rel="nofollow"><code>StableDiffusionPipeline</code></a>）</li> <li>基于文本和输入图像的引导生成（如 <a href="https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/img2img" rel="nofollow"><code>StableDiffusionImg2ImgPipeline</code></a> 和 <a href="https://huggingface.co/docs/diffusers/main/en/api/pipelines/pix2pix" rel="nofollow"><code>StableDiffusionInstructPix2PixPipeline</code></a>）</li> <li>类别条件图像生成模型(如 <a href="https://huggingface.co/docs/diffusers/main/en/api/pipe" rel="nofollow"><code>DiTPipeline</code></a>)</li></ul> <!> <p>定性评估通常涉及对生成图像的人工评判。评估维度包括构图质量、图文对齐度和空间关系等方面。标准化的提示词能为这些主观指标提供统一基准。DrawBench和PartiPrompts是常用的定性评估提示词数据集，分别由<a href="https://imagen.research.google/" rel="nofollow">Imagen</a>和<a href="https://parti.research.google/" rel="nofollow">Parti</a>团队提出。</p> <p>根据<a href="https://parti.research.google/" rel="nofollow">Parti官方网站</a>说明：</p> <blockquote><p>PartiPrompts (P2)是我们发布的包含1600多个英文提示词的丰富集合，可用于测量模型在不同类别和挑战维度上的能力。</p></blockquote> <p><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/parti-prompts.png" alt="parti-prompts"/></p> <p>PartiPrompts包含以下字段：</p> <ul><li>Prompt（提示词）</li> <li>Category（类别，如”抽象”、“世界知识”等）</li> <li>Challenge（难度等级，如”基础”、“复杂”、“文字与符号”等）</li></ul> <p>这些基准测试支持对不同图像生成模型进行并排人工对比评估。为此，🧨 Diffusers团队构建了<strong>Open Parti Prompts</strong>——一个基于Parti Prompts的社区驱动型定性评估基准，用于比较顶尖开源diffusion模型：</p> <ul><li><a href="https://huggingface.co/spaces/OpenGenAI/open-parti-prompts" rel="nofollow">Open Parti Prompts游戏</a>：展示10个parti提示词对应的4张生成图像，用户选择最符合提示的图片</li> <li><a href="https://huggingface.co/spaces/OpenGenAI/parti-prompts-leaderboard" rel="nofollow">Open Parti Prompts排行榜</a>：对比当前最优开源diffusion模型的性能榜单</li></ul> <p>为进行手动图像对比，我们演示如何使用<code>diffusers</code>处理部分PartiPrompts提示词。</p> <p>以下是从不同挑战维度（基础、复杂、语言结构、想象力、文字与符号）采样的提示词示例（使用<a href="https://huggingface.co/datasets/nateraw/parti-prompts" rel="nofollow">PartiPrompts作为数据集</a>）：</p> <!> <p>现在我们可以使用Stable Diffusion（<a href="https://huggingface.co/CompVis/stable-diffusion-v1-4" rel="nofollow">v1-4 checkpoint</a>）生成这些提示词对应的图像：</p> <!> <p><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/parti-prompts-14.png" alt="parti-prompts-14"/></p> <p>我们也可以通过设置<code>num_images_per_prompt</code>参数来比较同一提示词生成的不同图像。使用不同检查点(<a href="https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5" rel="nofollow">v1-5</a>)运行相同流程后，结果如下：</p> <p><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/parti-prompts-15.png" alt="parti-prompts-15"/></p> <p>当使用多个待评估模型为所有提示词生成若干图像后，这些结果将提交给人类评估员进行打分。有关DrawBench和PartiPrompts基准测试的更多细节，请参阅各自的论文。</p> <blockquote class="tip"><p>在模型训练过程中查看推理样本有助于评估训练进度。我们的<a href="https://github.com/huggingface/diffusers/tree/main/examples/" rel="nofollow">训练脚本</a>支持此功能，并额外提供TensorBoard和Weights & Biases日志记录功能。</p></blockquote> <!> <p>本节将指导您如何评估三种不同的扩散流程，使用以下指标：</p> <ul><li>CLIP分数</li> <li>CLIP方向相似度</li> <li>FID（弗雷歇起始距离）</li></ul> <!> <p><a href="https://huggingface.co/papers/2104.08718" rel="nofollow">CLIP分数</a>用于衡量图像-标题对的匹配程度。CLIP分数越高表明匹配度越高🔼。该分数是对”匹配度”这一定性概念的量化测量，也可以理解为图像与标题之间的语义相似度。研究发现CLIP分数与人类判断具有高度相关性。</p> <p>首先加载<code>StableDiffusionPipeline</code>：</p> <!> <p>使用多个提示词生成图像：</p> <!> <p>然后计算CLIP分数：</p> <!> <p>上述示例中，我们为每个提示生成一张图像。如果为每个提示生成多张图像，则需要计算每个提示生成图像的平均分数。</p> <p>当需要比较两个兼容<code>StableDiffusionPipeline</code>的检查点时，应在调用管道时传入生成器。首先使用<a href="https://huggingface.co/CompVis/stable-diffusion-v1-4" rel="nofollow">v1-4 Stable Diffusion检查点</a>以固定种子生成图像：</p> <!> <p>然后加载<a href="https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5" rel="nofollow">v1-5检查点</a>生成图像：</p> <!> <p>最后比较两者的CLIP分数：</p> <!> <p>结果表明<a href="https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5" rel="nofollow">v1-5</a>检查点性能优于前代。但需注意，我们用于计算CLIP分数的提示词数量较少。实际评估时应使用更多样化且数量更大的提示词集。</p> <blockquote class="warning"><p>该分数存在固有局限性：训练数据中的标题是从网络爬取，并提取自图片关联的<code>alt</code>等标签。这些描述未必符合人类描述图像的方式，因此我们需要人工”设计”部分提示词。</p></blockquote> <!> <p>这种情况下，生成管道同时接受输入图像和文本提示作为条件。以<code>StableDiffusionInstructPix2PixPipeline</code>为例，该管道接收编辑指令作为输入提示，并接受待编辑的输入图像。</p> <p>示例图示：</p> <p><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/edit-instruction.png" alt="编辑指令"/></p> <p>评估此类模型的策略之一是测量两幅图像间变化的连贯性（通过<a href="https://huggingface.co/docs/transformers/model_doc/clip" rel="nofollow">CLIP</a>定义）中两个图像之间的变化与两个图像描述之间的变化的一致性（如论文<a href="https://huggingface.co/papers/2108.00946" rel="nofollow">《CLIP-Guided Domain Adaptation of Image Generators》</a>所示）。这被称为“<strong>CLIP方向相似度</strong>”。</p> <ul><li><strong>描述1</strong>对应输入图像（图像1），即待编辑的图像。</li> <li><strong>描述2</strong>对应编辑后的图像（图像2），应反映编辑指令。</li></ul> <p>以下是示意图：</p> <p><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/edit-consistency.png" alt="edit-consistency"/></p> <p>我们准备了一个小型数据集来实现该指标。首先加载数据集：</p> <!> <!> <p>数据字段说明：</p> <ul><li><code>input</code>：与<code>image</code>对应的原始描述。</li> <li><code>edit</code>：编辑指令。</li> <li><code>output</code>：反映<code>edit</code>指令的修改后描述。</li></ul> <p>查看一个样本：</p> <!> <!> <p>对应的图像：</p> <!> <p><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/edit-dataset.png" alt="edit-dataset"/></p> <p>我们将根据编辑指令修改数据集中的图像，并计算方向相似度。</p> <p>首先加载<code>StableDiffusionInstructPix2PixPipeline</code>：</p> <!> <p>执行编辑操作：</p> <!> <p>为测量方向相似度，我们首先加载CLIP的图像和文本编码器：</p> <!> <p>注意我们使用的是特定CLIP检查点——<code>openai/clip-vit-large-patch14</code>，因为Stable Diffusion预训练正是基于此CLIP变体。详见<a href="https://huggingface.co/docs/transformers/model_doc/clip" rel="nofollow">文档</a>。</p> <p>接着准备计算方向相似度的PyTorch <code>nn.Module</code>：</p> <!> <p>现在让我们使用<code>DirectionalSimilarity</code>模块：</p> <!> <p>与CLIP分数类似，CLIP方向相似度数值越高越好。</p> <p>需要注意的是，<code>StableDiffusionInstructPix2PixPipeline</code>提供了两个控制参数<code>image_guidance_scale</code>和<code>guidance_scale</code>来调节最终编辑图像的质量。建议您尝试调整这两个参数，观察它们对方向相似度的影响。</p> <p>我们可以扩展这个度量标准来评估原始图像与编辑版本的相似度，只需计算<code>F.cosine_similarity(img_feat_two, img_feat_one)</code>。对于这类编辑任务，我们仍希望尽可能保留图像的主要语义特征（即保持较高的相似度分数）。</p> <p>该度量方法同样适用于类似流程，例如<a href="https://huggingface.co/docs/diffusers/main/en/api/pipelines/pix2pix_zero#diffusers.StableDiffusionPix2PixZeroPipeline" rel="nofollow"><code>StableDiffusionPix2PixZeroPipeline</code></a>。</p> <blockquote class="tip"><p>CLIP分数和CLIP方向相似度都依赖CLIP模型，可能导致评估结果存在偏差。</p></blockquote> <p><strong><em>扩展IS、FID（后文讨论）或KID等指标存在困难</em></strong>，当被评估模型是在大型图文数据集（如<a href="https://laion.ai/blog/laion-5b/" rel="nofollow">LAION-5B数据集</a>）上预训练时。因为这些指标的底层都使用了在ImageNet-1k数据集上预训练的InceptionNet来提取图像特征。Stable Diffusion的预训练数据集与InceptionNet的预训练数据集可能重叠有限，因此不适合作为特征提取器。</p> <p><strong><em>上述指标更适合评估类别条件模型</em></strong>，例如<a href="https://huggingface.co/docs/diffusers/main/en/api/pipelines/dit" rel="nofollow">DiT</a>。该模型是在ImageNet-1k类别条件下预训练的。
	这是9篇文档中的第8部分。</p> <!> <p>基于类别的生成模型通常是在带有类别标签的数据集（如<a href="https://huggingface.co/datasets/imagenet-1k" rel="nofollow">ImageNet-1k</a>）上进行预训练的。评估这些模型的常用指标包括Fréchet Inception Distance（FID）、Kernel Inception Distance（KID）和Inception Score（IS）。本文档重点介绍FID（<a href="https://huggingface.co/papers/1706.08500" rel="nofollow">Heusel等人</a>），并展示如何使用<a href="https://huggingface.co/docs/diffusers/api/pipelines/dit" rel="nofollow"><code>DiTPipeline</code></a>计算该指标，该管道底层使用了<a href="https://huggingface.co/papers/2212.09748" rel="nofollow">DiT模型</a>。</p> <p>FID旨在衡量两组图像数据集的相似程度。根据<a href="https://mmgeneration.readthedocs.io/en/latest/quick_run.html#fid" rel="nofollow">此资源</a>：</p> <blockquote><p>Fréchet Inception Distance是衡量两组图像数据集相似度的指标。研究表明其与人类对视觉质量的主观判断高度相关，因此最常用于评估生成对抗网络（GAN）生成样本的质量。FID通过计算Inception网络特征表示所拟合的两个高斯分布之间的Fréchet距离来实现。</p></blockquote> <p>这两个数据集本质上是真实图像数据集和生成图像数据集（本例中为人工生成的图像）。FID通常基于两个大型数据集计算，但本文档将使用两个小型数据集进行演示。</p> <p>首先下载ImageNet-1k训练集中的部分图像：</p> <!> <!> <p>这些是来自以下ImageNet-1k类别的10张图像：“cassette_player”、“chain_saw”（2张）、“church”、“gas_pump”（3张）、“parachute”（2张）和”tench”。</p> <p align="center"><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/real-images.png" alt="真实图像"/><br/> <em>真实图像</em></p> <p>加载图像后，我们对其进行轻量级预处理以便用于FID计算：</p> <!> <p>我们现在加载<a href="https://huggingface.co/docs/diffusers/api/pipelines/dit" rel="nofollow"><code>DiTPipeline</code></a>来生成基于上述类别的条件图像。</p> <!> <p>现在，我们可以使用<a href="https://torchmetrics.readthedocs.io/" rel="nofollow"><code>torchmetrics</code></a>计算FID分数。</p> <!> <p>FID分数越低越好。以下因素会影响FID结果：</p> <ul><li>图像数量（包括真实图像和生成图像）</li> <li>扩散过程中引入的随机性</li> <li>扩散过程的推理步数</li> <li>扩散过程中使用的调度器</li></ul> <p>对于最后两点，最佳实践是使用不同的随机种子和推理步数进行多次评估，然后报告平均结果。</p> <blockquote class="warning"><p>FID结果往往具有脆弱性，因为它依赖于许多因素：</p> <ul><li>计算过程中使用的特定Inception模型</li> <li>计算实现的准确性</li> <li>图像格式（PNG和JPG的起点不同）</li></ul> <p>需要注意的是，FID通常在比较相似实验时最有用，但除非作者仔细公开FID测量代码，否则很难复现论文结果。</p> <p>这些注意事项同样适用于其他相关指标，如KID和IS。</p></blockquote> <p>最后，让我们可视化检查这些<code>fake_images</code>。</p> <p align="center"><img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/fake-images.png" alt="生成图像"/><br/> <em>生成图像示例</em></p> <!> <p></p>`,1);function nl(E,Q){D(Q,!1),q(()=>{new URLSearchParams(window.location.search).get("fw")}),v();var e=ll();S("1sfanq2",R=>{var N=$();H(N,"content",O),F(R,N)});var n=l(P(e),2);z(n,{containerStyle:"float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"});var p=l(n,2);a(p,{title:"Diffusion模型评估指南",local:"diffusion模型评估指南",headingTag:"h1"});var t=l(p,14);a(t,{title:"评估场景",local:"评估场景",headingTag:"h2"});var i=l(t,6);a(i,{title:"定性评估",local:"定性评估",headingTag:"h2"});var o=l(i,22);s(o,{code:"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",highlighted:`<span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset

	<span class="hljs-comment"># prompts = load_dataset("nateraw/parti-prompts", split="train")</span>
	<span class="hljs-comment"># prompts = prompts.shuffle()</span>
	<span class="hljs-comment"># sample_prompts = [prompts[i]["Prompt"] for i in range(5)]</span>

	<span class="hljs-comment"># Fixing these sample prompts in the interest of reproducibility.</span>
	sample_prompts = [
	<span class="hljs-string">"a corgi"</span>,
	<span class="hljs-string">"a hot air balloon with a yin-yang symbol, with the moon visible in the daytime sky"</span>,
	<span class="hljs-string">"a car with no windows"</span>,
	<span class="hljs-string">"a cube made of porcupine"</span>,
	<span class="hljs-string">'The saying "BE EXCELLENT TO EACH OTHER" written on a red brick wall with a graffiti image of a green alien wearing a tuxedo. A yellow fire hydrant is on a sidewalk in the foreground.'</span>,
	]`,lang:"python",wrap:!1});var c=l(o,4);s(c,{code:"aW1wb3J0JTIwdG9yY2glMEElMEFzZWVkJTIwJTNEJTIwMCUwQWdlbmVyYXRvciUyMCUzRCUyMHRvcmNoLm1hbnVhbF9zZWVkKHNlZWQpJTBBJTBBaW1hZ2VzJTIwJTNEJTIwc2RfcGlwZWxpbmUoc2FtcGxlX3Byb21wdHMlMkMlMjBudW1faW1hZ2VzX3Blcl9wcm9tcHQlM0QxJTJDJTIwZ2VuZXJhdG9yJTNEZ2VuZXJhdG9yKS5pbWFnZXM=",highlighted:`<span class="hljs-keyword">import</span> torch

	seed = <span class="hljs-number">0</span>
	generator = torch.manual_seed(seed)

	images = sd_pipeline(sample_prompts, num_images_per_prompt=<span class="hljs-number">1</span>, generator=generator).images`,lang:"python",wrap:!1});var M=l(c,12);a(M,{title:"定量评估",local:"定量评估",headingTag:"h2"});var J=l(M,6);a(J,{title:"文本引导图像生成",local:"文本引导图像生成",headingTag:"h3"});var d=l(J,6);s(d,{code:"ZnJvbSUyMGRpZmZ1c2VycyUyMGltcG9ydCUyMFN0YWJsZURpZmZ1c2lvblBpcGVsaW5lJTBBaW1wb3J0JTIwdG9yY2glMEElMEFtb2RlbF9ja3B0JTIwJTNEJTIwJTIyQ29tcFZpcyUyRnN0YWJsZS1kaWZmdXNpb24tdjEtNCUyMiUwQXNkX3BpcGVsaW5lJTIwJTNEJTIwU3RhYmxlRGlmZnVzaW9uUGlwZWxpbmUuZnJvbV9wcmV0cmFpbmVkKG1vZGVsX2NrcHQlMkMlMjB0b3JjaF9kdHlwZSUzRHRvcmNoLmZsb2F0MTYpLnRvKCUyMmN1ZGElMjIp",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> StableDiffusionPipeline
	<span class="hljs-keyword">import</span> torch

	model_ckpt = <span class="hljs-string">"CompVis/stable-diffusion-v1-4"</span>
	sd_pipeline = StableDiffusionPipeline.from_pretrained(model_ckpt, torch_dtype=torch.float16).to(<span class="hljs-string">"cuda"</span>)`,lang:"python",wrap:!1});var r=l(d,4);s(r,{code:"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",highlighted:`prompts = [
	<span class="hljs-string">"a photo of an astronaut riding a horse on mars"</span>,
	<span class="hljs-string">"A high tech solarpunk utopia in the Amazon rainforest"</span>,
	<span class="hljs-string">"A pikachu fine dining with a view to the Eiffel Tower"</span>,
	<span class="hljs-string">"A mecha robot in a favela in expressionist style"</span>,
	<span class="hljs-string">"an insect robot preparing a delicious meal"</span>,
	<span class="hljs-string">"A small cabin on top of a snowy mountain in the style of Disney, artstation"</span>,
	]

	images = sd_pipeline(prompts, num_images_per_prompt=<span class="hljs-number">1</span>, output_type=<span class="hljs-string">"np"</span>).images

	<span class="hljs-built_in">print</span>(images.shape)
	<span class="hljs-comment"># (6, 512, 512, 3)</span>`,lang:"python",wrap:!1});var y=l(r,4);s(y,{code:"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",highlighted:`<span class="hljs-keyword">from</span> torchmetrics.functional.multimodal <span class="hljs-keyword">import</span> clip_score
	<span class="hljs-keyword">from</span> functools <span class="hljs-keyword">import</span> partial

	clip_score_fn = partial(clip_score, model_name_or_path=<span class="hljs-string">"openai/clip-vit-base-patch16"</span>)

	<span class="hljs-keyword">def</span> <span class="hljs-title function_">calculate_clip_score</span>(<span class="hljs-params">images, prompts</span>):
	images_int = (images * <span class="hljs-number">255</span>).astype(<span class="hljs-string">"uint8"</span>)
	clip_score = clip_score_fn(torch.from_numpy(images_int).permute(<span class="hljs-number">0</span>, <span class="hljs-number">3</span>, <span class="hljs-number">1</span>, <span class="hljs-number">2</span>), prompts).detach()
	<span class="hljs-keyword">return</span> <span class="hljs-built_in">round</span>(<span class="hljs-built_in">float</span>(clip_score), <span class="hljs-number">4</span>)

	sd_clip_score = calculate_clip_score(images, prompts)
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"CLIP分数: <span class="hljs-subst">{sd_clip_score}</span>"</span>)
	<span class="hljs-comment"># CLIP分数: 35.7038</span>`,lang:"python",wrap:!1});var h=l(y,6);s(h,{code:"c2VlZCUyMCUzRCUyMDAlMEFnZW5lcmF0b3IlMjAlM0QlMjB0b3JjaC5tYW51YWxfc2VlZChzZWVkKSUwQSUwQWltYWdlcyUyMCUzRCUyMHNkX3BpcGVsaW5lKHByb21wdHMlMkMlMjBudW1faW1hZ2VzX3Blcl9wcm9tcHQlM0QxJTJDJTIwZ2VuZXJhdG9yJTNEZ2VuZXJhdG9yJTJDJTIwb3V0cHV0X3R5cGUlM0QlMjJucCUyMikuaW1hZ2Vz",highlighted:`seed = <span class="hljs-number">0</span>
	generator = torch.manual_seed(seed)

	images = sd_pipeline(prompts, num_images_per_prompt=<span class="hljs-number">1</span>, generator=generator, output_type=<span class="hljs-string">"np"</span>).images`,lang:"python",wrap:!1});var j=l(h,4);s(j,{code:"bW9kZWxfY2twdF8xXzUlMjAlM0QlMjAlMjJzdGFibGUtZGlmZnVzaW9uLXYxLTUlMkZzdGFibGUtZGlmZnVzaW9uLXYxLTUlMjIlMEFzZF9waXBlbGluZV8xXzUlMjAlM0QlMjBTdGFibGVEaWZmdXNpb25QaXBlbGluZS5mcm9tX3ByZXRyYWluZWQobW9kZWxfY2twdF8xXzUlMkMlMjB0b3JjaF9kdHlwZSUzRHRvcmNoLmZsb2F0MTYpLnRvKCUyMmN1ZGElMjIpJTBBJTBBaW1hZ2VzXzFfNSUyMCUzRCUyMHNkX3BpcGVsaW5lXzFfNShwcm9tcHRzJTJDJTIwbnVtX2ltYWdlc19wZXJfcHJvbXB0JTNEMSUyQyUyMGdlbmVyYXRvciUzRGdlbmVyYXRvciUyQyUyMG91dHB1dF90eXBlJTNEJTIybnAlMjIpLmltYWdlcw==",highlighted:`model_ckpt_1_5 = <span class="hljs-string">"stable-diffusion-v1-5/stable-diffusion-v1-5"</span>
	sd_pipeline_1_5 = StableDiffusionPipeline.from_pretrained(model_ckpt_1_5, torch_dtype=torch.float16).to(<span class="hljs-string">"cuda"</span>)

	images_1_5 = sd_pipeline_1_5(prompts, num_images_per_prompt=<span class="hljs-number">1</span>, generator=generator, output_type=<span class="hljs-string">"np"</span>).images`,lang:"python",wrap:!1});var m=l(j,4);s(m,{code:"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",highlighted:`sd_clip_score_1_4 = calculate_clip_score(images, prompts)
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"v-1-4版本的CLIP分数: <span class="hljs-subst">{sd_clip_score_1_4}</span>"</span>)
	<span class="hljs-comment"># v-1-4版本的CLIP分数: 34.9102</span>

	sd_clip_score_1_5 = calculate_clip_score(images_1_5, prompts)
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"v-1-5版本的CLIP分数: <span class="hljs-subst">{sd_clip_score_1_5}</span>"</span>)
	<span class="hljs-comment"># v-1-5版本的CLIP分数: 36.2137</span>`,lang:"python",wrap:!1});var T=l(m,6);a(T,{title:"图像条件式文本生成图像",local:"图像条件式文本生成图像",headingTag:"h3"});var U=l(T,18);s(U,{code:"ZnJvbSUyMGRhdGFzZXRzJTIwaW1wb3J0JTIwbG9hZF9kYXRhc2V0JTBBJTBBZGF0YXNldCUyMCUzRCUyMGxvYWRfZGF0YXNldCglMjJzYXlha3BhdWwlMkZpbnN0cnVjdHBpeDJwaXgtZGVtbyUyMiUyQyUyMHNwbGl0JTNEJTIydHJhaW4lMjIpJTBBZGF0YXNldC5mZWF0dXJlcw==",highlighted:`<span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset

	dataset = load_dataset(<span class="hljs-string">"sayakpaul/instructpix2pix-demo"</span>, split=<span class="hljs-string">"train"</span>)
	dataset.features`,lang:"python",wrap:!1});var w=l(U,2);s(w,{code:"JTdCJ2lucHV0JyUzQSUyMFZhbHVlKGR0eXBlJTNEJ3N0cmluZyclMkMlMjBpZCUzRE5vbmUpJTJDJTBBJTIwJ2VkaXQnJTNBJTIwVmFsdWUoZHR5cGUlM0Qnc3RyaW5nJyUyQyUyMGlkJTNETm9uZSklMkMlMEElMjAnb3V0cHV0JyUzQSUyMFZhbHVlKGR0eXBlJTNEJ3N0cmluZyclMkMlMjBpZCUzRE5vbmUpJTJDJTBBJTIwJ2ltYWdlJyUzQSUyMEltYWdlKGRlY29kZSUzRFRydWUlMkMlMjBpZCUzRE5vbmUpJTdE",highlighted:`{<span class="hljs-string">'input'</span>: Value(dtype=<span class="hljs-string">'string'</span>, <span class="hljs-built_in">id</span>=None),
	<span class="hljs-string">'edit'</span>: Value(dtype=<span class="hljs-string">'string'</span>, <span class="hljs-built_in">id</span>=None),
	<span class="hljs-string">'output'</span>: Value(dtype=<span class="hljs-string">'string'</span>, <span class="hljs-built_in">id</span>=None),
	<span class="hljs-string">'image'</span>: Image(decode=True, <span class="hljs-built_in">id</span>=None)}`,lang:"bash",wrap:!1});var u=l(w,8);s(u,{code:"aWR4JTIwJTNEJTIwMCUwQXByaW50KGYlMjJPcmlnaW5hbCUyMGNhcHRpb24lM0ElMjAlN0JkYXRhc2V0JTVCaWR4JTVEJTVCJ2lucHV0JyU1RCU3RCUyMiklMEFwcmludChmJTIyRWRpdCUyMGluc3RydWN0aW9uJTNBJTIwJTdCZGF0YXNldCU1QmlkeCU1RCU1QidlZGl0JyU1RCU3RCUyMiklMEFwcmludChmJTIyTW9kaWZpZWQlMjBjYXB0aW9uJTNBJTIwJTdCZGF0YXNldCU1QmlkeCU1RCU1QidvdXRwdXQnJTVEJTdEJTIyKQ==",highlighted:`idx = <span class="hljs-number">0</span>
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"Original caption: <span class="hljs-subst">{dataset[idx][<span class="hljs-string">'input'</span>]}</span>"</span>)
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"Edit instruction: <span class="hljs-subst">{dataset[idx][<span class="hljs-string">'edit'</span>]}</span>"</span>)
	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"Modified caption: <span class="hljs-subst">{dataset[idx][<span class="hljs-string">'output'</span>]}</span>"</span>)`,lang:"python",wrap:!1});var b=l(u,2);s(b,{code:"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",highlighted:`Original caption: 2. FAROE ISLANDS: An archipelago of 18 mountainous isles <span class="hljs-keyword">in</span> the North Atlantic Ocean between Norway and Iceland, the Faroe Islands has <span class="hljs-string">'everything you could hope for'</span>, according to Big 7 Travel. It boasts <span class="hljs-string">'crystal clear waterfalls, rocky cliffs that seem to jut out of nowhere and velvety green hills'</span>
	Edit instruction: make the isles all white marble
	Modified caption: 2. WHITE MARBLE ISLANDS: An archipelago of 18 mountainous white marble isles <span class="hljs-keyword">in</span> the North Atlantic Ocean between Norway and Iceland, the White Marble Islands has <span class="hljs-string">'everything you could hope for'</span>, according to Big 7 Travel. It boasts <span class="hljs-string">'crystal clear waterfalls, rocky cliffs that seem to jut out of nowhere and velvety green hills'</span>`,lang:"bash",wrap:!1});var g=l(b,4);s(g,{code:"ZGF0YXNldCU1QmlkeCU1RCU1QiUyMmltYWdlJTIyJTVE",highlighted:'dataset[idx][<span class="hljs-string">"image"</span>]',lang:"python",wrap:!1});var I=l(g,8);s(I,{code:"ZnJvbSUyMGRpZmZ1c2VycyUyMGltcG9ydCUyMFN0YWJsZURpZmZ1c2lvbkluc3RydWN0UGl4MlBpeFBpcGVsaW5lJTBBJTBBaW5zdHJ1Y3RfcGl4MnBpeF9waXBlbGluZSUyMCUzRCUyMFN0YWJsZURpZmZ1c2lvbkluc3RydWN0UGl4MlBpeFBpcGVsaW5lLmZyb21fcHJldHJhaW5lZCglMEElMjAlMjAlMjAlMjAlMjJ0aW1icm9va3MlMkZpbnN0cnVjdC1waXgycGl4JTIyJTJDJTIwdG9yY2hfZHR5cGUlM0R0b3JjaC5mbG9hdDE2JTBBKS50byglMjJjdWRhJTIyKQ==",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> StableDiffusionInstructPix2PixPipeline

	instruct_pix2pix_pipeline = StableDiffusionInstructPix2PixPipeline.from_pretrained(
	<span class="hljs-string">"timbrooks/instruct-pix2pix"</span>, torch_dtype=torch.float16
	).to(<span class="hljs-string">"cuda"</span>)`,lang:"python",wrap:!1});var f=l(I,4);s(f,{code:"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",highlighted:`<span class="hljs-keyword">import</span> numpy <span class="hljs-keyword">as</span> np


	<span class="hljs-keyword">def</span> <span class="hljs-title function_">edit_image</span>(<span class="hljs-params">input_image, instruction</span>):
	image = instruct_pix2pix_pipeline(
	instruction,
	image=input_image,
	output_type=<span class="hljs-string">"np"</span>,
	generator=generator,
	).images[<span class="hljs-number">0</span>]
	<span class="hljs-keyword">return</span> image

	input_images = []
	original_captions = []
	modified_captions = []
	edited_images = []

	<span class="hljs-keyword">for</span> idx <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-built_in">len</span>(dataset)):
	input_image = dataset[idx][<span class="hljs-string">"image"</span>]
	edit_instruction = dataset[idx][<span class="hljs-string">"edit"</span>]
	edited_image = edit_image(input_image, edit_instruction)

	input_images.append(np.array(input_image))
	original_captions.append(dataset[idx][<span class="hljs-string">"input"</span>])
	modified_captions.append(dataset[idx][<span class="hljs-string">"output"</span>])
	edited_images.append(edited_image)`,lang:"python",wrap:!1});var Z=l(f,4);s(Z,{code:"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",highlighted:`<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> (
	CLIPTokenizer,
	CLIPTextModelWithProjection,
	CLIPVisionModelWithProjection,
	CLIPImageProcessor,
	)

	clip_id = <span class="hljs-string">"openai/clip-vit-large-patch14"</span>
	tokenizer = CLIPTokenizer.from_pretrained(clip_id)
	text_encoder = CLIPTextModelWithProjection.from_pretrained(clip_id).to(<span class="hljs-string">"cuda"</span>)
	image_processor = CLIPImageProcessor.from_pretrained(clip_id)
	image_encoder = CLIPVisionModelWithProjection.from_pretrained(clip_id).to(<span class="hljs-string">"cuda"</span>)`,lang:"python",wrap:!1});var B=l(Z,6);s(B,{code:"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",highlighted:`<span class="hljs-keyword">import</span> torch.nn <span class="hljs-keyword">as</span> nn
	<span class="hljs-keyword">import</span> torch.nn.functional <span class="hljs-keyword">as</span> F


	<span class="hljs-keyword">class</span> <span class="hljs-title class_">DirectionalSimilarity</span>(nn.Module):
	<span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self, tokenizer, text_encoder, image_processor, image_encoder</span>):
	<span class="hljs-built_in">super</span>().__init__()
	<span class="hljs-variable language_">self</span>.tokenizer = tokenizer
	<span class="hljs-variable language_">self</span>.text_encoder = text_encoder
	<span class="hljs-variable language_">self</span>.image_processor = image_processor
	<span class="hljs-variable language_">self</span>.image_encoder = image_encoder

	<span class="hljs-keyword">def</span> <span class="hljs-title function_">preprocess_image</span>(<span class="hljs-params">self, image</span>):
	image = <span class="hljs-variable language_">self</span>.image_processor(image, return_tensors=<span class="hljs-string">"pt"</span>)[<span class="hljs-string">"pixel_values"</span>]
	<span class="hljs-keyword">return</span> {<span class="hljs-string">"pixel_values"</span>: image.to(<span class="hljs-string">"cuda"</span>)}

	<span class="hljs-keyword">def</span> <span class="hljs-title function_">tokenize_text</span>(<span class="hljs-params">self, text</span>):
	inputs = <span class="hljs-variable language_">self</span>.tokenizer(
	text,
	max_length=<span class="hljs-variable language_">self</span>.tokenizer.model_max_length,
	padding=<span class="hljs-string">"max_length"</span>,
	truncation=<span class="hljs-literal">True</span>,
	return_tensors=<span class="hljs-string">"pt"</span>,
	)
	<span class="hljs-keyword">return</span> {<span class="hljs-string">"input_ids"</span>: inputs.input_ids.to(<span class="hljs-string">"cuda"</span>)}

	<span class="hljs-keyword">def</span> <span class="hljs-title function_">encode_image</span>(<span class="hljs-params">self, image</span>):
	preprocessed_image = <span class="hljs-variable language_">self</span>.preprocess_image(image)
	image_features = <span class="hljs-variable language_">self</span>.image_encoder(**preprocessed_image).image_embeds
	image_features = image_features / image_features.norm(dim=<span class="hljs-number">1</span>, keepdim=<span class="hljs-literal">True</span>)
	<span class="hljs-keyword">return</span> image_features

	<span class="hljs-keyword">def</span> <span class="hljs-title function_">encode_text</span>(<span class="hljs-params">self, text</span>):
	tokenized_text = <span class="hljs-variable language_">self</span>.tokenize_text(text)
	text_features = <span class="hljs-variable language_">self</span>.text_encoder(**tokenized_text).text_embeds
	text_features = text_features / text_features.norm(dim=<span class="hljs-number">1</span>, keepdim=<span class="hljs-literal">True</span>)
	<span class="hljs-keyword">return</span> text_features

	<span class="hljs-keyword">def</span> <span class="hljs-title function_">compute_directional_similarity</span>(<span class="hljs-params">self, img_feat_one, img_feat_two, text_feat_one, text_feat_two</span>):
	sim_direction = F.cosine_similarity(img_feat_two - img_feat_one, text_feat_two - text_feat_one)
	<span class="hljs-keyword">return</span> sim_direction

	<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, image_one, image_two, caption_one, caption_two</span>):
	img_feat_one = <span class="hljs-variable language_">self</span>.encode_image(image_one)
	img_feat_two = <span class="hljs-variable language_">self</span>.encode_image(image_two)
	text_feat_one = <span class="hljs-variable language_">self</span>.encode_text(caption_one)
	text_feat_two = <span class="hljs-variable language_">self</span>.encode_text(caption_two)
	directional_similarity = <span class="hljs-variable language_">self</span>.compute_directional_similarity(
	img_feat_one, img_feat_two, text_feat_one, text_feat_two
	)
	<span class="hljs-keyword">return</span> directional_similarity`,lang:"python",wrap:!1});var W=l(B,4);s(W,{code:"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",highlighted:`dir_similarity = DirectionalSimilarity(tokenizer, text_encoder, image_processor, image_encoder)
	scores = []

	<span class="hljs-keyword">for</span> i <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-built_in">len</span>(input_images)):
	original_image = input_images[i]
	original_caption = original_captions[i]
	edited_image = edited_images[i]
	modified_caption = modified_captions[i]

	similarity_score = dir_similarity(original_image, edited_image, original_caption, modified_caption)
	scores.append(<span class="hljs-built_in">float</span>(similarity_score.detach().cpu()))

	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"CLIP方向相似度: <span class="hljs-subst">{np.mean(scores)}</span>"</span>)
	<span class="hljs-comment"># CLIP方向相似度: 0.0797976553440094</span>`,lang:"python",wrap:!1});var G=l(W,16);a(G,{title:"基于类别的图像生成",local:"基于类别的图像生成",headingTag:"h3"});var C=l(G,12);s(C,{code:"ZnJvbSUyMHppcGZpbGUlMjBpbXBvcnQlMjBaaXBGaWxlJTBBaW1wb3J0JTIwcmVxdWVzdHMlMEElMEElMEFkZWYlMjBkb3dubG9hZCh1cmwlMkMlMjBsb2NhbF9maWxlcGF0aCklM0ElMEElMjAlMjAlMjAlMjByJTIwJTNEJTIwcmVxdWVzdHMuZ2V0KHVybCklMEElMjAlMjAlMjAlMjB3aXRoJTIwb3Blbihsb2NhbF9maWxlcGF0aCUyQyUyMCUyMndiJTIyKSUyMGFzJTIwZiUzQSUwQSUyMCUyMCUyMCUyMCUyMCUyMCUyMCUyMGYud3JpdGUoci5jb250ZW50KSUwQSUyMCUyMCUyMCUyMHJldHVybiUyMGxvY2FsX2ZpbGVwYXRoJTBBJTBBZHVtbXlfZGF0YXNldF91cmwlMjAlM0QlMjAlMjJodHRwcyUzQSUyRiUyRmhmLmNvJTJGZGF0YXNldHMlMkZzYXlha3BhdWwlMkZzYW1wbGUtZGF0YXNldHMlMkZyZXNvbHZlJTJGbWFpbiUyRnNhbXBsZS1pbWFnZW5ldC1pbWFnZXMuemlwJTIyJTBBbG9jYWxfZmlsZXBhdGglMjAlM0QlMjBkb3dubG9hZChkdW1teV9kYXRhc2V0X3VybCUyQyUyMGR1bW15X2RhdGFzZXRfdXJsLnNwbGl0KCUyMiUyRiUyMiklNUItMSU1RCklMEElMEF3aXRoJTIwWmlwRmlsZShsb2NhbF9maWxlcGF0aCUyQyUyMCUyMnIlMjIpJTIwYXMlMjB6aXBwZXIlM0ElMEElMjAlMjAlMjAlMjB6aXBwZXIuZXh0cmFjdGFsbCglMjIuJTIyKQ==",highlighted:`<span class="hljs-keyword">from</span> zipfile <span class="hljs-keyword">import</span> ZipFile
	<span class="hljs-keyword">import</span> requests


	<span class="hljs-keyword">def</span> <span class="hljs-title function_">download</span>(<span class="hljs-params">url, local_filepath</span>):
	r = requests.get(url)
	<span class="hljs-keyword">with</span> <span class="hljs-built_in">open</span>(local_filepath, <span class="hljs-string">"wb"</span>) <span class="hljs-keyword">as</span> f:
	f.write(r.content)
	<span class="hljs-keyword">return</span> local_filepath

	dummy_dataset_url = <span class="hljs-string">"https://hf.co/datasets/sayakpaul/sample-datasets/resolve/main/sample-imagenet-images.zip"</span>
	local_filepath = download(dummy_dataset_url, dummy_dataset_url.split(<span class="hljs-string">"/"</span>)[-<span class="hljs-number">1</span>])

	<span class="hljs-keyword">with</span> ZipFile(local_filepath, <span class="hljs-string">"r"</span>) <span class="hljs-keyword">as</span> zipper:
	zipper.extractall(<span class="hljs-string">"."</span>)`,lang:"python",wrap:!1});var _=l(C,2);s(_,{code:"ZnJvbSUyMFBJTCUyMGltcG9ydCUyMEltYWdlJTBBaW1wb3J0JTIwb3MlMEFpbXBvcnQlMjBudW1weSUyMGFzJTIwbnAlMEElMEFkYXRhc2V0X3BhdGglMjAlM0QlMjAlMjJzYW1wbGUtaW1hZ2VuZXQtaW1hZ2VzJTIyJTBBaW1hZ2VfcGF0aHMlMjAlM0QlMjBzb3J0ZWQoJTVCb3MucGF0aC5qb2luKGRhdGFzZXRfcGF0aCUyQyUyMHgpJTIwZm9yJTIweCUyMGluJTIwb3MubGlzdGRpcihkYXRhc2V0X3BhdGgpJTVEKSUwQSUwQXJlYWxfaW1hZ2VzJTIwJTNEJTIwJTVCbnAuYXJyYXkoSW1hZ2Uub3BlbihwYXRoKS5jb252ZXJ0KCUyMlJHQiUyMikpJTIwZm9yJTIwcGF0aCUyMGluJTIwaW1hZ2VfcGF0aHMlNUQ=",highlighted:`<span class="hljs-keyword">from</span> PIL <span class="hljs-keyword">import</span> Image
	<span class="hljs-keyword">import</span> os
	<span class="hljs-keyword">import</span> numpy <span class="hljs-keyword">as</span> np

	dataset_path = <span class="hljs-string">"sample-imagenet-images"</span>
	image_paths = <span class="hljs-built_in">sorted</span>([os.path.join(dataset_path, x) <span class="hljs-keyword">for</span> x <span class="hljs-keyword">in</span> os.listdir(dataset_path)])

	real_images = [np.array(Image.<span class="hljs-built_in">open</span>(path).convert(<span class="hljs-string">"RGB"</span>)) <span class="hljs-keyword">for</span> path <span class="hljs-keyword">in</span> image_paths]`,lang:"python",wrap:!1});var V=l(_,8);s(V,{code:"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",highlighted:`<span class="hljs-keyword">from</span> torchvision.transforms <span class="hljs-keyword">import</span> functional <span class="hljs-keyword">as</span> F
	<span class="hljs-keyword">import</span> torch


	<span class="hljs-keyword">def</span> <span class="hljs-title function_">preprocess_image</span>(<span class="hljs-params">image</span>):
	image = torch.tensor(image).unsqueeze(<span class="hljs-number">0</span>)
	image = image.permute(<span class="hljs-number">0</span>, <span class="hljs-number">3</span>, <span class="hljs-number">1</span>, <span class="hljs-number">2</span>) / <span class="hljs-number">255.0</span>
	<span class="hljs-keyword">return</span> F.center_crop(image, (<span class="hljs-number">256</span>, <span class="hljs-number">256</span>))

	real_images = torch.stack([dit_pipeline.preprocess_image(image) <span class="hljs-keyword">for</span> image <span class="hljs-keyword">in</span> real_images])
	<span class="hljs-built_in">print</span>(real_images.shape)
	<span class="hljs-comment"># torch.Size([10, 3, 256, 256])</span>`,lang:"python",wrap:!1});var X=l(V,4);s(X,{code:"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",highlighted:`<span class="hljs-keyword">from</span> diffusers <span class="hljs-keyword">import</span> DiTPipeline, DPMSolverMultistepScheduler

	dit_pipeline = DiTPipeline.from_pretrained(<span class="hljs-string">"facebook/DiT-XL-2-256"</span>, torch_dtype=torch.float16)
	dit_pipeline.scheduler = DPMSolverMultistepScheduler.from_config(dit_pipeline.scheduler.config)
	dit_pipeline = dit_pipeline.to(<span class="hljs-string">"cuda"</span>)

	seed = <span class="hljs-number">0</span>
	generator = torch.manual_seed(seed)


	words = [
	<span class="hljs-string">"cassette player"</span>,
	<span class="hljs-string">"chainsaw"</span>,
	<span class="hljs-string">"chainsaw"</span>,
	<span class="hljs-string">"church"</span>,
	<span class="hljs-string">"gas pump"</span>,
	<span class="hljs-string">"gas pump"</span>,
	<span class="hljs-string">"gas pump"</span>,
	<span class="hljs-string">"parachute"</span>,
	<span class="hljs-string">"parachute"</span>,
	<span class="hljs-string">"tench"</span>,
	]

	class_ids = dit_pipeline.get_label_ids(words)
	output = dit_pipeline(class_labels=class_ids, generator=generator, output_type=<span class="hljs-string">"np"</span>)

	fake_images = output.images
	fake_images = torch.tensor(fake_images)
	fake_images = fake_images.permute(<span class="hljs-number">0</span>, <span class="hljs-number">3</span>, <span class="hljs-number">1</span>, <span class="hljs-number">2</span>)
	<span class="hljs-built_in">print</span>(fake_images.shape)
	<span class="hljs-comment"># torch.Size([10, 3, 256, 256])</span>`,lang:"python",wrap:!1});var k=l(X,4);s(k,{code:"ZnJvbSUyMHRvcmNobWV0cmljcy5pbWFnZS5maWQlMjBpbXBvcnQlMjBGcmVjaGV0SW5jZXB0aW9uRGlzdGFuY2UlMEElMEFmaWQlMjAlM0QlMjBGcmVjaGV0SW5jZXB0aW9uRGlzdGFuY2Uobm9ybWFsaXplJTNEVHJ1ZSklMEFmaWQudXBkYXRlKHJlYWxfaW1hZ2VzJTJDJTIwcmVhbCUzRFRydWUpJTBBZmlkLnVwZGF0ZShmYWtlX2ltYWdlcyUyQyUyMHJlYWwlM0RGYWxzZSklMEElMEFwcmludChmJTIyRklEJUU1JTg4JTg2JUU2JTk1JUIwJTNBJTIwJTdCZmxvYXQoZmlkLmNvbXB1dGUoKSklN0QlMjIpJTBBJTIzJTIwRklEJUU1JTg4JTg2JUU2JTk1JUIwJTNBJTIwMTc3LjcxNDcyMTY3OTY4NzU=",highlighted:`<span class="hljs-keyword">from</span> torchmetrics.image.fid <span class="hljs-keyword">import</span> FrechetInceptionDistance

	fid = FrechetInceptionDistance(normalize=<span class="hljs-literal">True</span>)
	fid.update(real_images, real=<span class="hljs-literal">True</span>)
	fid.update(fake_images, real=<span class="hljs-literal">False</span>)

	<span class="hljs-built_in">print</span>(<span class="hljs-string">f"FID分数: <span class="hljs-subst">{<span class="hljs-built_in">float</span>(fid.compute())}</span>"</span>)
	<span class="hljs-comment"># FID分数: 177.7147216796875</span>`,lang:"python",wrap:!1});var A=l(k,14);x(A,{source:"https://github.com/huggingface/diffusers/blob/main/docs/source/zh/conceptual/evaluation.md"}),K(2),F(E,e),L()}export{nl as component};

Xet Storage Details

Size:: 59.3 kB
Xet hash:: 74312bed8130499b365b33f7355285dfcfbbb74e719f761c5b2f9074d12ca849

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.