depth-map-generation

深度图生成

使用 each::sense 从任意图像生成精准的深度图。这项技能可从 2D 图像中提取深度信息，适用于 3D 效果、视差动画、VR/AR 应用、计算摄影等。

功能

• 单目深度估计：从单张图像中提取深度
• 人像深度图：为人像摄影效果提供精准深度
• 风景深度：提取全景和风景图像的场景深度
• 产品深度：为电商生成适用于 3D 的深度图
• 建筑深度：分析建筑和室内场景的深度
• 3D 视差效果：为 Ken Burns 风格动画提供深度数据
• VR/AR 深度：为沉浸式体验实时估计深度
• 立体图像生成：将 2D 图像转换为立体 3D
• 焦点堆栈：基于深度选择焦平面
• 背景虚化：支持深度感知的焦外虚化和模糊效果

快速开始

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this photo"}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/photo.jpg"],
    "mode": "max"
  }'

深度图输出格式

格式	描述	用途
灰度	8 位深度（白色 = 近，黑色 = 远）	通用用途、可视化
反向灰度	8 位深度（黑色 = 近，白色 = 远）	兼容部分 3D 软件
归一化	0-1 范围的深度值	机器学习流水线
度量	估算现实世界距离	AR/VR、机器人

用例示例

1. 从照片生成深度图

从任意图像中提取深度的基本方法。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this image. Output as grayscale where white represents closer objects and black represents distant objects."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/scene.jpg"],
    "mode": "max"
  }'

2. 人像深度图

从人像照片中提取精准深度，用于焦外虚化效果和 3D 人像。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Create a high-precision depth map from this portrait photo. Focus on accurate edge detection around the subject, especially hair and facial features. I need this for applying realistic depth-of-field effects."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/portrait.jpg"],
    "mode": "max"
  }'

3. 风景深度图

从风景和户外场景中生成深度，并提供更大的深度范围。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this landscape photo. The scene has foreground elements, mid-ground terrain, and distant mountains. Capture the full depth range from near to far with good separation between depth layers."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/landscape.jpg"],
    "mode": "max"
  }'

4. 用于 3D 效果的产品深度

为产品图像创建深度图，以支持 3D 查看体验。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Extract a depth map from this product photo. I need accurate depth information to create a 3D interactive view for an e-commerce website. Focus on capturing the product shape and surface details."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/product-sneaker.jpg"],
    "mode": "max"
  }'

5. 建筑深度图

从建筑和室内照片中生成深度，用于可视化。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Create a depth map from this interior architecture photo. Capture the spatial relationships between walls, furniture, and architectural elements. I need this for a virtual tour with depth-based transitions."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/interior.jpg"],
    "mode": "max"
  }'

6. 创建 3D 视差效果

生成适用于创建视差动画和 Ken Burns 效果的深度图。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this photo that I will use for a 3D parallax animation. I need clear depth separation between foreground, midground, and background elements. The depth should be smooth with distinct layers for a compelling parallax effect."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/scene-for-parallax.jpg"],
    "mode": "max"
  }'

7. VR/AR 深度估计

创建适用于虚拟现实和增强现实应用的深度图。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this room photo for AR/VR use. I need metric depth estimation that accurately represents real-world distances. This will be used for placing virtual objects in an augmented reality application."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/room.jpg"],
    "mode": "max"
  }'

8. 立体图像生成

使用深度估计将 2D 图像转换为立体 3D。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this photo and use it to create a stereoscopic 3D image pair (left and right eye views). The stereo effect should be subtle enough for comfortable viewing but noticeable enough to create depth perception."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/photo-for-stereo.jpg"],
    "mode": "max"
  }'

9. 焦点堆栈深度

生成深度图，用于计算摄影焦点堆栈和全景深合成。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Create a depth map from this macro/close-up photo. I need precise depth information to identify focus planes for computational focus stacking. Each depth layer should be clearly defined so I can select which areas should be in focus."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/macro-photo.jpg"],
    "mode": "max"
  }'

10. 深度感知背景虚化

生成深度信息，以应用逼真的焦外虚化和背景模糊效果。

curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this photo for applying depth-aware background blur. The subject in the foreground should be clearly separated from the background. I need accurate edge detection so the blur transition looks natural, similar to a portrait mode effect."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/photo-for-blur.jpg"],
    "mode": "max"
  }'

多轮深度处理

使用 session_id 优化深度图或处理多张相关图像：

# Initial depth estimation
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this photo"}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/scene.jpg"],
    "session_id": "depth-project-001"
  }'

# Refine the depth map
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "The depth map looks good but can you enhance the edge detection around the main subject? The boundaries are a bit fuzzy."}],
    "model": "eachsense/beta",
    "stream": true,
    "session_id": "depth-project-001"
  }'

# Apply the depth map for an effect
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Now use this depth map to create a 3D parallax video animation with subtle camera movement"}],
    "model": "eachsense/beta",
    "stream": true,
    "session_id": "depth-project-001"
  }'

批量深度处理

处理多张图像，以获得一致的深度估计：

# Process first image
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate a depth map from this product photo. I will be sending more product images that need consistent depth estimation."}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/product1.jpg"],
    "session_id": "product-depth-batch",
    "mode": "eco"
  }'

# Process second image with same settings
curl -X POST https://eachsense-agent.core.eachlabs.run/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: $EACHLABS_API_KEY" \
  -H "Accept: text/event-stream" \
  -d '{
    "messages": [{"role": "user", "content": "Generate depth map for this product using the same approach as before"}],
    "model": "eachsense/beta",
    "stream": true,
    "image_urls": ["https://example.com/product2.jpg"],
    "session_id": "product-depth-batch",
    "mode": "eco"
  }'

模式选择

模式	最适用场景	速度	质量
max	最终制作深度图、VR/AR 应用、专业合成	较慢	精度最高
eco	快速预览、批量处理、原型制作	较快	质量良好

**建议：**当深度精度至关重要时（VR/AR、3D 转换、专业合成），使用 max 模式。快速迭代和批量处理时使用 eco 模式。

最佳实践

输入图像质量

• 分辨率：输入分辨率越高，生成的深度图细节越丰富
• 光照：均匀的光照有助于精准估计深度
• 对比度：物体之间对比清晰，有助于区分深度
• 对焦：清晰的图像有助于更准确地检测深度图中的边缘

深度图应用

• 视差效果：使用 3-5 个不同的深度层，以获得最佳效果
• 焦外虚化/模糊：确保主体边缘清晰干净，使模糊过渡自然
• 3D 转换：提供场景尺度信息，便于估算度量深度
• VR/AR：请求度量深度，以准确估算现实世界距离

提示词技巧

1. 指定输出格式：灰度、归一化或度量深度
2. 描述场景：帮助模型理解空间关系
3. 说明使用场景：不同应用适合不同的深度特征
4. 提出边缘质量要求：说明需要清晰还是平滑的深度过渡

错误处理

错误	原因	解决方案
Failed to create prediction: HTTP 422	余额不足	在 eachlabs.ai 充值
图像加载失败	图像 URL 无效或无法访问	确认图像 URL 可公开访问
超时	图像过于复杂或分辨率过高	将客户端超时设为至少 10 分钟
深度输出质量较低	输入图像质量较差	使用分辨率更高、光线更好的源图像

技术说明

• 客户端超时：对于复杂的深度估算，将 HTTP 客户端超时设为至少 10 分钟
• 图像格式：支持 JPEG、PNG、WebP 输入图像
• 输出格式：深度图通常以灰度 PNG 图像的形式输出
• 深度范围：默认使用相对深度（0-1）；也可按需提供度量深度

相关技能

• each-sense - 核心 API 文档
• image-to-3d - 根据图像生成完整的 3D 模型
• image-editing - 对图像应用基于深度的效果
• video-generation - 根据深度图创建视差视频