test-repo/docs/.vitepress/theme/components/appendix/image-gen-intro/UNetDenoiseDemo.vue

<!--
  UNetDenoiseDemo.vue
  UNet 去噪过程演示组件

  用途：
  展示 UNet/DiT 如何从噪声中逐步恢复图像，理解扩散模型的核心去噪机制。

  交互功能：
  - 单步/自动播放去噪过程
  - 可视化噪声预测
  - 展示不同时间步的预测结果
  - 对比有/无文本引导的生成
-->
<template>
  <div class="unet-demo">
    <el-card shadow="never">
      <template #header>
        <div class="header-controls">
          <span class="title">🧠 UNet 去噪模型</span>
          <div class="controls">
            <el-button-group>
              <el-button @click="stepBackward" :disabled="currentStep <= 0">
                <el-icon><ArrowLeft /></el-icon>
              </el-button>
              <el-button @click="togglePlay">
                <el-icon v-if="isPlaying"><VideoPause /></el-icon>
                <el-icon v-else><VideoPlay /></el-icon>
              </el-button>
              <el-button @click="stepForward" :disabled="currentStep >= totalSteps">
                <el-icon><ArrowRight /></el-icon>
              </el-button>
            </el-button-group>
            <el-button @click="reset">重置</el-button>
          </div>
        </div>
      </template>

      <div class="demo-content">
        <!-- 主展示区 -->
        <div class="main-display">
          <div class="display-section">
            <div class="section-label">当前噪声图像 (Noisy Image)</div>
            <canvas
              ref="noisyCanvas"
              width="256"
              height="256"
              class="display-canvas"
            />
            <div class="timestep-info">
              <el-tag type="info">Timestep: {{ currentStep }} / {{ totalSteps }}</el-tag>
              <el-tag :type="getNoiseLevelType()">噪声强度: {{ getNoiseLevel() }}%</el-tag>
            </div>
          </div>

          <div class="arrow-section">
            <el-icon :size="24"><ArrowRight /></el-icon>
            <div class="model-box">
              <div class="model-name">UNet / DiT</div>
              <div class="model-desc">预测噪声</div>
            </div>
            <el-icon :size="24"><ArrowRight /></el-icon>
          </div>

          <div class="display-section">
            <div class="section-label">预测的噪声 (Predicted Noise)</div>
            <canvas
              ref="noiseCanvas"
              width="256"
              height="256"
              class="display-canvas noise-preview"
            />
            <div class="noise-stats">
              <el-tag size="small" type="warning">噪声估计</el-tag>
            </div>
          </div>

          <div class="arrow-section">
            <el-icon :size="24"><ArrowRight /></el-icon>
            <div class="operation-box">
              <div class="op-name">减法</div>
              <div class="op-formula">x - ε</div>
            </div>
            <el-icon :size="24"><ArrowRight /></el-icon>
          </div>

          <div class="display-section">
            <div class="section-label">去噪结果 (Denoised)</div>
            <canvas
              ref="denoisedCanvas"
              width="256"
              height="256"
              class="display-canvas"
            />
            <div class="progress-info">
              <el-progress
                :percentage="(currentStep / totalSteps) * 100"
                :status="currentStep === totalSteps ? 'success' : ''"
              />
            </div>
          </div>
        </div>

        <!-- 时间轴 -->
        <div class="timeline-section">
          <div class="timeline-label">去噪时间轴</div>
          <el-slider
            v-model="currentStep"
            :min="0"
            :max="totalSteps"
            :step="1"
            show-stops
            :marks="marks"
            @input="updateDisplay"
          />
        </div>

        <!-- 对比模式 -->
        <div class="compare-section">
          <el-switch
            v-model="showComparison"
            active-text="显示对比 (有/无文本引导)"
          />
          <div v-if="showComparison" class="compare-display">
            <div class="compare-item">
              <div class="compare-label">无引导 (Unconditional)</div>
              <canvas
                ref="uncondCanvas"
                width="200"
                height="200"
                class="compare-canvas"
              />
            </div>
            <div class="compare-item">
              <div class="compare-label">有引导 (CFG Scale=7.5)</div>
              <canvas
                ref="condCanvas"
                width="200"
                height="200"
                class="compare-canvas"
              />
            </div>
          </div>
        </div>
      </div>

      <div class="info-box">
        <p>
          <span class="icon">💡</span>
          <strong>去噪原理：</strong>
          UNet 学习预测图像中的噪声，然后用原图减去预测的噪声，得到更清晰的结果。重复这个过程，直到从纯噪声恢复出清晰图像。
        </p>
      </div>
    </el-card>
  </div>
</template>

<script setup>
import { ref, onMounted, onUnmounted } from 'vue'
import { ArrowRight, ArrowLeft, VideoPlay, VideoPause } from '@element-plus/icons-vue'

const noisyCanvas = ref(null)
const noiseCanvas = ref(null)
const denoisedCanvas = ref(null)
const uncondCanvas = ref(null)
const condCanvas = ref(null)

const currentStep = ref(0)
const totalSteps = 20
const isPlaying = ref(false)
const showComparison = ref(false)

const marks = {
  0: '纯噪声',
  10: '中期',
  20: '清晰图'
}

let animationId = null

// 生成目标图像（简化版）
const generateTargetImage = () => {
  const canvas = document.createElement('canvas')
  canvas.width = 256
  canvas.height = 256
  const ctx = canvas.getContext('2d')

  // 绘制简单的目标图案
  const gradient = ctx.createLinearGradient(0, 0, 256, 256)
  gradient.addColorStop(0, '#667eea')
  gradient.addColorStop(1, '#764ba2')
  ctx.fillStyle = gradient
  ctx.fillRect(0, 0, 256, 256)

  // 添加一些形状
  ctx.fillStyle = 'rgba(255, 255, 255, 0.3)'
  for (let i = 0; i < 5; i++) {
    ctx.beginPath()
    ctx.arc(50 + i * 40, 100 + (i % 2) * 50, 30, 0, Math.PI * 2)
    ctx.fill()
  }

  return ctx.getImageData(0, 0, 256, 256)
}

const targetImage = generateTargetImage()

// 生成噪声
const generateNoise = (width, height, intensity) => {
  const data = new Uint8ClampedArray(width * height * 4)
  for (let i = 0; i < data.length; i += 4) {
    const noise = (Math.random() - 0.5) * intensity * 255
    data[i] = 128 + noise
    data[i + 1] = 128 + noise
    data[i + 2] = 128 + noise
    data[i + 3] = 255
  }
  return new ImageData(data, width, height)
}

// 混合图像和噪声
const blendWithNoise = (imageData, noiseRatio) => {
  const result = new Uint8ClampedArray(imageData.data)
  for (let i = 0; i < result.length; i += 4) {
    const noise = (Math.random() - 0.5) * noiseRatio * 255
    result[i] = Math.max(0, Math.min(255, imageData.data[i] * (1 - noiseRatio) + 128 * noiseRatio + noise))
    result[i + 1] = Math.max(0, Math.min(255, imageData.data[i + 1] * (1 - noiseRatio) + 128 * noiseRatio + noise))
    result[i + 2] = Math.max(0, Math.min(255, imageData.data[i + 2] * (1 - noiseRatio) + 128 * noiseRatio + noise))
  }
  return new ImageData(result, imageData.width, imageData.height)
}

// 预测噪声（简化模拟）
const predictNoise = (width, height, step) => {
  const noiseRatio = 1 - (step / totalSteps)
  return generateNoise(width, height, noiseRatio * 0.5)
}

// 去噪
const denoise = (noisyData, noiseData, step) => {
  const result = new Uint8ClampedArray(noisyData.data)
  const denoiseStrength = 0.1 + (step / totalSteps) * 0.4

  for (let i = 0; i < result.length; i += 4) {
    // 模拟：从噪声图像中减去预测的噪声
    const targetR = targetImage.data[i]
    const targetG = targetImage.data[i + 1]
    const targetB = targetImage.data[i + 2]

    const currentR = noisyData.data[i]
    const currentG = noisyData.data[i + 1]
    const currentB = noisyData.data[i + 2]

    result[i] = currentR + (targetR - currentR) * denoiseStrength
    result[i + 1] = currentG + (targetG - currentG) * denoiseStrength
    result[i + 2] = currentB + (targetB - currentB) * denoiseStrength
  }

  return new ImageData(result, noisyData.width, noisyData.height)
}

// 更新显示
const updateDisplay = () => {
  const step = currentStep.value
  const noiseRatio = 1 - (step / totalSteps)

  // 绘制噪声图像
  const noisyCtx = noisyCanvas.value.getContext('2d')
  const noisyData = blendWithNoise(targetImage, noiseRatio)
  noisyCtx.putImageData(noisyData, 0, 0)

  // 绘制预测的噪声
  const noiseCtx = noiseCanvas.value.getContext('2d')
  const noiseData = predictNoise(256, 256, step)
  noiseCtx.putImageData(noiseData, 0, 0)

  // 绘制去噪结果
  const denoisedCtx = denoisedCanvas.value.getContext('2d')
  const denoisedData = denoise(noisyData, noiseData, step)
  denoisedCtx.putImageData(denoisedData, 0, 0)

  // 更新对比图
  if (showComparison.value && uncondCanvas.value && condCanvas.value) {
    // 无条件生成（更多噪声残留）
    const uncondCtx = uncondCanvas.value.getContext('2d')
    const uncondData = blendWithNoise(targetImage, noiseRatio * 0.3)
    uncondCtx.putImageData(uncondData, 0, 0)

    // 有条件生成（更清晰）
    const condCtx = condCanvas.value.getContext('2d')
    condCtx.putImageData(denoisedData, 0, 0)
  }
}

const getNoiseLevel = () => {
  return Math.round((1 - currentStep.value / totalSteps) * 100)
}

const getNoiseLevelType = () => {
  const level = getNoiseLevel()
  if (level > 70) return 'danger'
  if (level > 30) return 'warning'
  return 'success'
}

const stepForward = () => {
  if (currentStep.value < totalSteps) {
    currentStep.value++
    updateDisplay()
  }
}

const stepBackward = () => {
  if (currentStep.value > 0) {
    currentStep.value--
    updateDisplay()
  }
}

const togglePlay = () => {
  if (isPlaying.value) {
    stopAnimation()
  } else {
    startAnimation()
  }
}

const startAnimation = () => {
  isPlaying.value = true
  const animate = () => {
    if (!isPlaying.value) return

    if (currentStep.value >= totalSteps) {
      currentStep.value = 0
    } else {
      currentStep.value++
    }
    updateDisplay()

    animationId = setTimeout(() => {
      requestAnimationFrame(animate)
    }, 200)
  }
  animate()
}

const stopAnimation = () => {
  isPlaying.value = false
  if (animationId) {
    clearTimeout(animationId)
    animationId = null
  }
}

const reset = () => {
  stopAnimation()
  currentStep.value = 0
  updateDisplay()
}

onMounted(updateDisplay)
onUnmounted(stopAnimation)
</script>

<style scoped>
.unet-demo {
  margin: 1rem 0;
}

.header-controls {
  display: flex;
  justify-content: space-between;
  align-items: center;
  flex-wrap: wrap;
  gap: 12px;
}

.title {
  font-weight: 600;
}

.controls {
  display: flex;
  gap: 8px;
}

.demo-content {
  display: flex;
  flex-direction: column;
  gap: 24px;
}

.main-display {
  display: flex;
  align-items: center;
  justify-content: center;
  gap: 16px;
  flex-wrap: wrap;
  padding: 16px 0;
}

.display-section {
  display: flex;
  flex-direction: column;
  align-items: center;
  gap: 8px;
}

.section-label {
  font-size: 0.875rem;
  font-weight: 500;
  color: var(--vp-c-text-2);
}

.display-canvas {
  width: 200px;
  height: 200px;
  background: var(--vp-c-bg-mute);
  border-radius: 8px;
  border: 2px solid var(--vp-c-divider);
}

.noise-preview {
  filter: grayscale(100%);
}

.timestep-info {
  display: flex;
  gap: 8px;
}

.arrow-section {
  display: flex;
  flex-direction: column;
  align-items: center;
  gap: 8px;
  color: var(--vp-c-text-3);
}

.model-box,
.operation-box {
  background: var(--vp-c-bg-soft);
  border: 1px solid var(--vp-c-divider);
  border-radius: 8px;
  padding: 12px 16px;
  text-align: center;
  min-width: 80px;
}

.model-name,
.op-name {
  font-weight: 600;
  font-size: 0.875rem;
}

.model-desc,
.op-formula {
  font-size: 0.75rem;
  color: var(--vp-c-text-3);
  margin-top: 4px;
}

.progress-info {
  width: 100%;
  max-width: 200px;
}

.timeline-section {
  padding: 16px;
  background: var(--vp-c-bg-soft);
  border-radius: 8px;
}

.timeline-label {
  font-weight: 500;
  margin-bottom: 12px;
}

.compare-section {
  padding: 16px;
  background: var(--vp-c-bg-soft);
  border-radius: 8px;
}

.compare-display {
  display: flex;
  justify-content: center;
  gap: 24px;
  margin-top: 16px;
  flex-wrap: wrap;
}

.compare-item {
  display: flex;
  flex-direction: column;
  align-items: center;
  gap: 8px;
}

.compare-label {
  font-size: 0.875rem;
  color: var(--vp-c-text-2);
}

.compare-canvas {
  width: 150px;
  height: 150px;
  background: var(--vp-c-bg-mute);
  border-radius: 8px;
  border: 2px solid var(--vp-c-divider);
}

.info-box {
  margin-top: 16px;
  padding: 12px;
  background: var(--vp-c-bg-mute);
  border-radius: 6px;
  font-size: 0.9rem;
  line-height: 1.6;
}

.icon {
  font-size: 1.2em;
}

@media (max-width: 768px) {
  .main-display {
    flex-direction: column;
  }

  .arrow-section {
    transform: rotate(90deg);
    margin: 8px 0;
  }
}
</style>