Multimodal Embeddings API

Multimodal Embeddings API 会根据您提供的输入生成向量，输入可以包含图片、文本和视频数据的组合。然后，嵌入向量可用于后续任务，例如图片分类或视频内容审核。

如需了解额外的概念信息，请参阅多模态嵌入。

支持的模型：

模型 代码
Embeddings for Multimodal multimodalembedding@001

示例语法

用于发送多模嵌入 API 请求的语法。

curl

curl -X POST \
-H "Authorization: Bearer $(gcloud auth print-access-token)" \
-H "Content-Type: application/json" \

https://${LOCATION}-aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/${LOCATION}/publishers/google/models/${MODEL_ID}:predict \
-d '{
"instances": [
  ...
],
}'

Python

from vertexai.vision_models import MultiModalEmbeddingModel

model = MultiModalEmbeddingModel.from_pretrained("multimodalembedding")
model.get_embeddings(...)

参数列表

如需了解实现详情，请参阅示例。

请求正文

{
  "instances": [
    {
      "text": string,
      "image": {
        // Union field can be only one of the following:
        "bytesBase64Encoded": string,
        "gcsUri": string,
        // End of list of possible types for union field.
        "mimeType": string
      },
      "video": {
        // Union field can be only one of the following:
        "bytesBase64Encoded": string,
        "gcsUri": string,
        // End of list of possible types for union field.
        "videoSegmentConfig": {
          "startOffsetSec": integer,
          "endOffsetSec": integer,
          "intervalSec": integer
        }
      },
      "parameters": {
        "dimension": integer
      }
    }
  ]
}
参数

image

可选：Image。

要为其生成嵌入的图片。

text

可选：String。

要为其生成嵌入的文本。

video

可选：Video。

要为其生成嵌入的视频片段。

dimension

可选：Int。

响应中包含的嵌入维度。仅适用于文本和图片输入。接受的值：128、256、512 或 1408。

图片

参数

bytesBase64Encoded

可选：String。

以 base64 字符串编码的图片字节。必须是 bytesBase64Encoded 或 gcsUri 中的一个。

gcsUri

可选。String

要执行嵌入操作的图片的 Cloud Storage 位置。bytesBase64Encoded 或 gcsUri 之一。

mimeType

可选。String

图片内容的 MIME 类型。支持的值：image/jpeg 和 image/png。

视频

参数

bytesBase64Encoded

可选：String。

以 base64 字符串编码的视频字节。bytesBase64Encoded 或 gcsUri 之一。

gcsUri

可选：String。

要执行嵌入操作的视频的 Cloud Storage 位置。bytesBase64Encoded 或 gcsUri 之一。

videoSegmentConfig

可选：VideoSegmentConfig。

视频片段配置。
VideoSegmentConfig
参数

startOffsetSec

可选：Int。

视频片段的开始偏移量（以秒为单位）。如果未指定，则使用 max(0, endOffsetSec - 120) 进行计算。

endOffsetSec

可选：Int。

视频片段的结束偏移量（以秒为单位）。如果未指定，则使用 min(video length, startOffSec + 120) 进行计算。如果同时指定了 startOffSec 和 endOffSec，则 endOffsetSec 会调整为 min(startOffsetSec+120, endOffsetSec)。

intervalSec

可选。Int

要生成嵌入的视频的间隔。interval_sec 的最小值为 4。如果间隔小于 4，则返回 InvalidArgumentError。间隔的最大值没有限制。不过，如果间隔时间超过 min(video length, 120s)，则会影响生成的嵌入的质量。默认值：16。

响应正文

{
  "predictions": [
    {
      "textEmbedding": [
        float,
        // array of 128, 256, 512, or 1408 float values
        float
      ],
      "imageEmbedding": [
        float,
        // array of 128, 256, 512, or 1408 float values
        float
      ],
      "videoEmbeddings": [
        {
          "startOffsetSec": integer,
          "endOffsetSec": integer,
          "embedding": [
            float,
            // array of 1408 float values
            float
          ]
        }
      ]
    }
  ],
  "deployedModelId": string
}
响应元素 说明
imageEmbedding 128、256、512 或 1408 维度的浮点数列表。
textEmbedding 128、256、512 或 1408 维度的浮点数列表。
videoEmbeddings 1408 维度的浮点数列表，其中包含为其生成嵌入的视频片段的开始时间和结束时间（以秒为单位）。

示例

基本用例

从图片生成嵌入

使用以下示例为图片生成嵌入。

REST

在使用任何请求数据之前，请先进行以下替换：

  • LOCATION：您的项目的区域。例如 us-central1、europe-west2 或 asia-northeast3。如需查看可用区域的列表，请参阅 Vertex AI 上的生成式 AI 位置。
  • PROJECT_ID：您的 Google Cloud 项目 ID。
  • TEXT：要获取嵌入的目标文本。例如：a cat。
  • B64_ENCODED_IMG：要获取嵌入的目标图片。图片必须指定为 base64 编码的字节字符串。

HTTP 方法和网址：

POST https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict

请求 JSON 正文： 

{
  "instances": [
    {
      "text": "TEXT",
      "image": {
        "bytesBase64Encoded": "B64_ENCODED_IMG"
      }
    }
  ]
}

如需发送请求，请选择以下方式之一：

curl

将请求正文保存在名为 request.json 的文件中，然后执行以下命令： 

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict"

PowerShell

将请求正文保存在名为 request.json 的文件中，然后执行以下命令： 

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict" | Select-Object -Expand Content
模型返回的嵌入是 1408 个浮点矢量。以下示例响应会缩短。 
{
  "predictions": [
    {
      "textEmbedding": [
        0.010477379,
        -0.00399621,
        0.00576670747,
        [...]
        -0.00823613815,
        -0.0169572588,
        -0.00472954148
      ],
      "imageEmbedding": [
        0.00262696808,
        -0.00198890246,
        0.0152047109,
        -0.0103145819,
        [...]
        0.0324628279,
        0.0284924973,
        0.011650892,
        -0.00452344026
      ]
    }
  ],
  "deployedModelId": "DEPLOYED_MODEL_ID"
}

Python

如需了解如何安装或更新 Vertex AI SDK for Python，请参阅安装 Vertex AI SDK for Python。 如需了解详情，请参阅 Python API 参考文档。 

import vertexai
from vertexai.vision_models import Image, MultiModalEmbeddingModel

# TODO(developer): Update & uncomment line below
# PROJECT_ID = "your-project-id"
vertexai.init(project=PROJECT_ID, location="us-central1")

model = MultiModalEmbeddingModel.from_pretrained("multimodalembedding@001")
image = Image.load_from_file(
    "gs://cloud-samples-data/vertex-ai/llm/prompts/landmark1.png"
)

embeddings = model.get_embeddings(
    image=image,
    contextual_text="Colosseum",
    dimension=1408,
)
print(f"Image Embedding: {embeddings.image_embedding}")
print(f"Text Embedding: {embeddings.text_embedding}")
# Example response:
# Image Embedding: [-0.0123147098, 0.0727171078, ...]
# Text Embedding: [0.00230263756, 0.0278981831, ...]

Node.js

在尝试此示例之前，请按照《Vertex AI 快速入门：使用客户端库》中的 Node.js 设置说明执行操作。 如需了解详情，请参阅 Vertex AI Node.js API 参考文档。

如需向 Vertex AI 进行身份验证，请设置应用默认凭证。 如需了解详情，请参阅为本地开发环境设置身份验证。 

/**
 * TODO(developer): Uncomment these variables before running the sample.\
 * (Not necessary if passing values as arguments)
 */
// const project = 'YOUR_PROJECT_ID';
// const location = 'YOUR_PROJECT_LOCATION';
// const baseImagePath = 'YOUR_BASE_IMAGE_PATH';
// const textPrompt = 'YOUR_TEXT_PROMPT';
const aiplatform = require('@google-cloud/aiplatform');

// Imports the Google Cloud Prediction service client
const {PredictionServiceClient} = aiplatform.v1;

// Import the helper module for converting arbitrary protobuf.Value objects.
const {helpers} = aiplatform;

// Specifies the location of the api endpoint
const clientOptions = {
  apiEndpoint: 'us-central1-aiplatform.googleapis.com',
};
const publisher = 'google';
const model = 'multimodalembedding@001';

// Instantiates a client
const predictionServiceClient = new PredictionServiceClient(clientOptions);

async function predictImageFromImageAndText() {
  // Configure the parent resource
  const endpoint = `projects/${project}/locations/${location}/publishers/${publisher}/models/${model}`;

  const fs = require('fs');
  const imageFile = fs.readFileSync(baseImagePath);

  // Convert the image data to a Buffer and base64 encode it.
  const encodedImage = Buffer.from(imageFile).toString('base64');

  const prompt = {
    text: textPrompt,
    image: {
      bytesBase64Encoded: encodedImage,
    },
  };
  const instanceValue = helpers.toValue(prompt);
  const instances = [instanceValue];

  const parameter = {
    sampleCount: 1,
  };
  const parameters = helpers.toValue(parameter);

  const request = {
    endpoint,
    instances,
    parameters,
  };

  // Predict request
  const [response] = await predictionServiceClient.predict(request);
  console.log('Get image embedding response');
  const predictions = response.predictions;
  console.log('\tPredictions :');
  for (const prediction of predictions) {
    console.log(`\t\tPrediction : ${JSON.stringify(prediction)}`);
  }
}

await predictImageFromImageAndText();

Java

在尝试此示例之前，请按照《Vertex AI 快速入门：使用客户端库》中的 Java 设置说明执行操作。 如需了解详情，请参阅 Vertex AI Java API 参考文档。

如需向 Vertex AI 进行身份验证，请设置应用默认凭证。 如需了解详情，请参阅为本地开发环境设置身份验证。 


import com.google.cloud.aiplatform.v1beta1.EndpointName;
import com.google.cloud.aiplatform.v1beta1.PredictResponse;
import com.google.cloud.aiplatform.v1beta1.PredictionServiceClient;
import com.google.cloud.aiplatform.v1beta1.PredictionServiceSettings;
import com.google.gson.Gson;
import com.google.gson.JsonObject;
import com.google.protobuf.InvalidProtocolBufferException;
import com.google.protobuf.Value;
import com.google.protobuf.util.JsonFormat;
import java.io.IOException;
import java.nio.charset.StandardCharsets;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.ArrayList;
import java.util.Base64;
import java.util.HashMap;
import java.util.List;
import java.util.Map;

public class PredictImageFromImageAndTextSample {

  public static void main(String[] args) throws IOException {
    // TODO(developer): Replace this variable before running the sample.
    String project = "YOUR_PROJECT_ID";
    String textPrompt = "YOUR_TEXT_PROMPT";
    String baseImagePath = "YOUR_BASE_IMAGE_PATH";

    // Learn how to use text prompts to update an image:
    // https://cloud.google.com/vertex-ai/docs/generative-ai/image/edit-images
    Map<String, Object> parameters = new HashMap<String, Object>();
    parameters.put("sampleCount", 1);

    String location = "us-central1";
    String publisher = "google";
    String model = "multimodalembedding@001";

    predictImageFromImageAndText(
        project, location, publisher, model, textPrompt, baseImagePath, parameters);
  }

  // Update images using text prompts
  public static void predictImageFromImageAndText(
      String project,
      String location,
      String publisher,
      String model,
      String textPrompt,
      String baseImagePath,
      Map<String, Object> parameters)
      throws IOException {
    final String endpoint = String.format("%s-aiplatform.googleapis.com:443", location);
    final PredictionServiceSettings predictionServiceSettings =
        PredictionServiceSettings.newBuilder().setEndpoint(endpoint).build();

    // Initialize client that will be used to send requests. This client only needs to be created
    // once, and can be reused for multiple requests.
    try (PredictionServiceClient predictionServiceClient =
        PredictionServiceClient.create(predictionServiceSettings)) {
      final EndpointName endpointName =
          EndpointName.ofProjectLocationPublisherModelName(project, location, publisher, model);

      // Convert the image to Base64
      byte[] imageData = Base64.getEncoder().encode(Files.readAllBytes(Paths.get(baseImagePath)));
      String encodedImage = new String(imageData, StandardCharsets.UTF_8);

      JsonObject jsonInstance = new JsonObject();
      jsonInstance.addProperty("text", textPrompt);
      JsonObject jsonImage = new JsonObject();
      jsonImage.addProperty("bytesBase64Encoded", encodedImage);
      jsonInstance.add("image", jsonImage);

      Value instanceValue = stringToValue(jsonInstance.toString());
      List<Value> instances = new ArrayList<>();
      instances.add(instanceValue);

      Gson gson = new Gson();
      String gsonString = gson.toJson(parameters);
      Value parameterValue = stringToValue(gsonString);

      PredictResponse predictResponse =
          predictionServiceClient.predict(endpointName, instances, parameterValue);
      System.out.println("Predict Response");
      System.out.println(predictResponse);
      for (Value prediction : predictResponse.getPredictionsList()) {
        System.out.format("\tPrediction: %s\n", prediction);
      }
    }
  }

  // Convert a Json string to a protobuf.Value
  static Value stringToValue(String value) throws InvalidProtocolBufferException {
    Value.Builder builder = Value.newBuilder();
    JsonFormat.parser().merge(value, builder);
    return builder.build();
  }
}

Go

在尝试此示例之前，请按照《Vertex AI 快速入门：使用客户端库》中的 Go 设置说明执行操作。 如需了解详情，请参阅 Vertex AI Go API 参考文档。

如需向 Vertex AI 进行身份验证，请设置应用默认凭证。 如需了解详情，请参阅为本地开发环境设置身份验证。 

import (
	"context"
	"encoding/json"
	"fmt"
	"io"

	aiplatform "cloud.google.com/go/aiplatform/apiv1beta1"
	aiplatformpb "cloud.google.com/go/aiplatform/apiv1beta1/aiplatformpb"
	"google.golang.org/api/option"
	"google.golang.org/protobuf/encoding/protojson"
	"google.golang.org/protobuf/types/known/structpb"
)

// generateForTextAndImage shows how to use the multimodal model to generate embeddings for
// text and image inputs.
func generateForTextAndImage(w io.Writer, project, location string) error {
	// location = "us-central1"
	ctx := context.Background()
	apiEndpoint := fmt.Sprintf("%s-aiplatform.googleapis.com:443", location)
	client, err := aiplatform.NewPredictionClient(ctx, option.WithEndpoint(apiEndpoint))
	if err != nil {
		return fmt.Errorf("failed to construct API client: %w", err)
	}
	defer client.Close()

	model := "multimodalembedding@001"
	endpoint := fmt.Sprintf("projects/%s/locations/%s/publishers/google/models/%s", project, location, model)

	// This is the input to the model's prediction call. For schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#request_body
	instance, err := structpb.NewValue(map[string]any{
		"image": map[string]any{
			// Image input can be provided either as a Google Cloud Storage URI or as
			// base64-encoded bytes using the "bytesBase64Encoded" field.
			"gcsUri": "gs://cloud-samples-data/vertex-ai/llm/prompts/landmark1.png",
		},
		"text": "Colosseum",
	})
	if err != nil {
		return fmt.Errorf("failed to construct request payload: %w", err)
	}

	req := &aiplatformpb.PredictRequest{
		Endpoint: endpoint,
		// The model supports only 1 instance per request.
		Instances: []*structpb.Value{instance},
	}

	resp, err := client.Predict(ctx, req)
	if err != nil {
		return fmt.Errorf("failed to generate embeddings: %w", err)
	}

	instanceEmbeddingsJson, err := protojson.Marshal(resp.GetPredictions()[0])
	if err != nil {
		return fmt.Errorf("failed to convert protobuf value to JSON: %w", err)
	}
	// For response schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#response-body
	var instanceEmbeddings struct {
		ImageEmbeddings []float32 `json:"imageEmbedding"`
		TextEmbeddings  []float32 `json:"textEmbedding"`
	}
	if err := json.Unmarshal(instanceEmbeddingsJson, &instanceEmbeddings); err != nil {
		return fmt.Errorf("failed to unmarshal JSON: %w", err)
	}

	imageEmbedding := instanceEmbeddings.ImageEmbeddings
	textEmbedding := instanceEmbeddings.TextEmbeddings

	fmt.Fprintf(w, "Text embedding (length=%d): %v\n", len(textEmbedding), textEmbedding)
	fmt.Fprintf(w, "Image embedding (length=%d): %v\n", len(imageEmbedding), imageEmbedding)
	// Example response:
	// Text embedding (length=1408): [0.0023026613 0.027898183 -0.011858357 ... ]
	// Image embedding (length=1408): [-0.012314269 0.07271844 0.00020170923 ... ]

	return nil
}

根据视频生成嵌入

使用以下示例生成视频内容的嵌入。

REST

以下示例使用位于 Cloud Storage 中的视频。您还可以使用 video.bytesBase64Encoded 字段提供视频的 base64 编码字符串表示形式。

在使用任何请求数据之前，请先进行以下替换：

  • LOCATION：您的项目的区域。例如 us-central1、europe-west2 或 asia-northeast3。如需查看可用区域的列表，请参阅 Vertex AI 上的生成式 AI 位置。
  • PROJECT_ID：您的 Google Cloud 项目 ID。
  • VIDEO_URI：要为其获取嵌入的目标视频的 Cloud Storage URI。 例如 gs://my-bucket/embeddings/supermarket-video.mp4。

    您还可以以 base64 编码的字节字符串形式提供视频：

    [...]
"video": {
  "bytesBase64Encoded": "B64_ENCODED_VIDEO"
}
[...]
  • videoSegmentConfig（START_SECOND、END_SECOND、INTERVAL_SECONDS）。可选。为其生成嵌入的特定视频片段（以秒为单位）。

    例如：

    [...]
"videoSegmentConfig": {
  "startOffsetSec": 10,
  "endOffsetSec": 60,
  "intervalSec": 10
}
[...]

    使用此配置可指定从 10 秒到 60 秒的视频数据，并为以下 10 秒的视频间隔生成嵌入：[10, 20), [20, 30), [30, 40), [40, 50), [50, 60)。 此视频间隔 ("intervalSec": 10) 属于标准视频嵌入模式，用户按标准模式价格费率计费。

    如果省略 videoSegmentConfig，则服务使用以下默认值： "videoSegmentConfig": { "startOffsetSec": 0, "endOffsetSec": 120, "intervalSec": 16 }。 此视频间隔 ("intervalSec": 16) 属于基本视频嵌入模式，用户按基本模式价格费率计费。

HTTP 方法和网址：

POST https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict

请求 JSON 正文： 

{
  "instances": [
    {
      "video": {
        "gcsUri": "VIDEO_URI",
        "videoSegmentConfig": {
          "startOffsetSec": START_SECOND,
          "endOffsetSec": END_SECOND,
          "intervalSec": INTERVAL_SECONDS
        }
      }
    }
  ]
}

如需发送请求，请选择以下方式之一：

curl

将请求正文保存在名为 request.json 的文件中，然后执行以下命令： 

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict"

PowerShell

将请求正文保存在名为 request.json 的文件中，然后执行以下命令： 

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict" | Select-Object -Expand Content
模型返回的嵌入是 1408 个浮点矢量。以下响应示例缩短了空间。

响应（7 秒视频，未指定 videoSegmentConfig）：

{
  "predictions": [
    {
      "videoEmbeddings": [
        {
          "endOffsetSec": 7,
          "embedding": [
            -0.0045467657,
            0.0258095954,
            0.0146885719,
            0.00945400633,
            [...]
            -0.0023291884,
            -0.00493789,
            0.00975185353,
            0.0168156829
          ],
          "startOffsetSec": 0
        }
      ]
    }
  ],
  "deployedModelId": "DEPLOYED_MODEL_ID"
}

响应（59 秒视频，采用以下视频片段配置："videoSegmentConfig": { "startOffsetSec": 0, "endOffsetSec": 60, "intervalSec": 10 }）：

{
  "predictions": [
    {
      "videoEmbeddings": [
        {
          "endOffsetSec": 10,
          "startOffsetSec": 0,
          "embedding": [
            -0.00683252793,
            0.0390476175,
            [...]
            0.00657121744,
            0.013023301
          ]
        },
        {
          "startOffsetSec": 10,
          "endOffsetSec": 20,
          "embedding": [
            -0.0104404651,
            0.0357737206,
            [...]
            0.00509833824,
            0.0131902946
          ]
        },
        {
          "startOffsetSec": 20,
          "embedding": [
            -0.0113538112,
            0.0305239167,
            [...]
            -0.00195809244,
            0.00941874553
          ],
          "endOffsetSec": 30
        },
        {
          "embedding": [
            -0.00299320649,
            0.0322436653,
            [...]
            -0.00993082579,
            0.00968887936
          ],
          "startOffsetSec": 30,
          "endOffsetSec": 40
        },
        {
          "endOffsetSec": 50,
          "startOffsetSec": 40,
          "embedding": [
            -0.00591270532,
            0.0368893594,
            [...]
            -0.00219071587,
            0.0042470959
          ]
        },
        {
          "embedding": [
            -0.00458270218,
            0.0368121453,
            [...]
            -0.00317760976,
            0.00595594104
          ],
          "endOffsetSec": 59,
          "startOffsetSec": 50
        }
      ]
    }
  ],
  "deployedModelId": "DEPLOYED_MODEL_ID"
}

Python

如需了解如何安装或更新 Vertex AI SDK for Python，请参阅安装 Vertex AI SDK for Python。 如需了解详情，请参阅 Python API 参考文档。 

import vertexai

from vertexai.vision_models import MultiModalEmbeddingModel, Video
from vertexai.vision_models import VideoSegmentConfig

# TODO(developer): Update & uncomment line below
# PROJECT_ID = "your-project-id"
vertexai.init(project=PROJECT_ID, location="us-central1")

model = MultiModalEmbeddingModel.from_pretrained("multimodalembedding@001")

embeddings = model.get_embeddings(
    video=Video.load_from_file(
        "gs://cloud-samples-data/vertex-ai-vision/highway_vehicles.mp4"
    ),
    video_segment_config=VideoSegmentConfig(end_offset_sec=1),
)

# Video Embeddings are segmented based on the video_segment_config.
print("Video Embeddings:")
for video_embedding in embeddings.video_embeddings:
    print(
        f"Video Segment: {video_embedding.start_offset_sec} - {video_embedding.end_offset_sec}"
    )
    print(f"Embedding: {video_embedding.embedding}")

# Example response:
# Video Embeddings:
# Video Segment: 0.0 - 1.0
# Embedding: [-0.0206376351, 0.0123456789, ...]

Go

在尝试此示例之前，请按照《Vertex AI 快速入门：使用客户端库》中的 Go 设置说明执行操作。 如需了解详情，请参阅 Vertex AI Go API 参考文档。

如需向 Vertex AI 进行身份验证，请设置应用默认凭证。 如需了解详情，请参阅为本地开发环境设置身份验证。 

import (
	"context"
	"encoding/json"
	"fmt"
	"io"
	"time"

	aiplatform "cloud.google.com/go/aiplatform/apiv1beta1"
	aiplatformpb "cloud.google.com/go/aiplatform/apiv1beta1/aiplatformpb"
	"google.golang.org/api/option"
	"google.golang.org/protobuf/encoding/protojson"
	"google.golang.org/protobuf/types/known/structpb"
)

// generateForVideo shows how to use the multimodal model to generate embeddings for video input.
func generateForVideo(w io.Writer, project, location string) error {
	// location = "us-central1"

	// The default context timeout may be not enough to process a video input.
	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
	defer cancel()

	apiEndpoint := fmt.Sprintf("%s-aiplatform.googleapis.com:443", location)
	client, err := aiplatform.NewPredictionClient(ctx, option.WithEndpoint(apiEndpoint))
	if err != nil {
		return fmt.Errorf("failed to construct API client: %w", err)
	}
	defer client.Close()

	model := "multimodalembedding@001"
	endpoint := fmt.Sprintf("projects/%s/locations/%s/publishers/google/models/%s", project, location, model)

	// This is the input to the model's prediction call. For schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#request_body
	instances, err := structpb.NewValue(map[string]any{
		"video": map[string]any{
			// Video input can be provided either as a Google Cloud Storage URI or as base64-encoded
			// bytes using the "bytesBase64Encoded" field.
			"gcsUri": "gs://cloud-samples-data/vertex-ai-vision/highway_vehicles.mp4",
			"videoSegmentConfig": map[string]any{
				"startOffsetSec": 1,
				"endOffsetSec":   5,
			},
		},
	})
	if err != nil {
		return fmt.Errorf("failed to construct request payload: %w", err)
	}

	req := &aiplatformpb.PredictRequest{
		Endpoint: endpoint,
		// The model supports only 1 instance per request.
		Instances: []*structpb.Value{instances},
	}
	resp, err := client.Predict(ctx, req)
	if err != nil {
		return fmt.Errorf("failed to generate embeddings: %w", err)
	}

	instanceEmbeddingsJson, err := protojson.Marshal(resp.GetPredictions()[0])
	if err != nil {
		return fmt.Errorf("failed to convert protobuf value to JSON: %w", err)
	}
	// For response schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#response-body
	var instanceEmbeddings struct {
		VideoEmbeddings []struct {
			Embedding      []float32 `json:"embedding"`
			StartOffsetSec float64   `json:"startOffsetSec"`
			EndOffsetSec   float64   `json:"endOffsetSec"`
		} `json:"videoEmbeddings"`
	}
	if err := json.Unmarshal(instanceEmbeddingsJson, &instanceEmbeddings); err != nil {
		return fmt.Errorf("failed to unmarshal json: %w", err)
	}
	// Get the embedding for our single video segment (`.videoEmbeddings` object has one entry per
	// each processed segment).
	videoEmbedding := instanceEmbeddings.VideoEmbeddings[0]

	fmt.Fprintf(w, "Video embedding (seconds: %.f-%.f; length=%d): %v\n",
		videoEmbedding.StartOffsetSec,
		videoEmbedding.EndOffsetSec,
		len(videoEmbedding.Embedding),
		videoEmbedding.Embedding,
	)
	// Example response:
	// Video embedding (seconds: 1-5; length=1408): [-0.016427778 0.032878537 -0.030755188 ... ]

	return nil
}

高级用例

使用以下示例获取视频、文本和图片内容的嵌入。

对于视频嵌入，您可以指定视频片段和嵌入密度。

REST

以下示例使用图片、文本和视频数据。您可以在请求正文中使用这些数据类型的任意组合。

此示例会使用 Cloud Storage 中的视频。您还可以使用 video.bytesBase64Encoded 字段提供视频的 base64 编码字符串表示形式。

在使用任何请求数据之前，请先进行以下替换：

  • LOCATION：您的项目的区域。例如 us-central1、europe-west2 或 asia-northeast3。如需查看可用区域的列表，请参阅 Vertex AI 上的生成式 AI 位置。
  • PROJECT_ID：您的 Google Cloud 项目 ID。
  • TEXT：要获取嵌入的目标文本。例如：a cat。
  • IMAGE_URI：要获取其嵌入的目标图片的 Cloud Storage URI。例如 gs://my-bucket/embeddings/supermarket-img.png。

    您还可以以 base64 编码的字节字符串形式提供图片：

    [...]
"image": {
  "bytesBase64Encoded": "B64_ENCODED_IMAGE"
}
[...]
  • VIDEO_URI：要为其获取嵌入的目标视频的 Cloud Storage URI。 例如 gs://my-bucket/embeddings/supermarket-video.mp4。

    您还可以以 base64 编码的字节字符串形式提供视频：

    [...]
"video": {
  "bytesBase64Encoded": "B64_ENCODED_VIDEO"
}
[...]
  • videoSegmentConfig（START_SECOND、END_SECOND、INTERVAL_SECONDS）。可选。为其生成嵌入的特定视频片段（以秒为单位）。

    例如：

    [...]
"videoSegmentConfig": {
  "startOffsetSec": 10,
  "endOffsetSec": 60,
  "intervalSec": 10
}
[...]

    使用此配置可指定从 10 秒到 60 秒的视频数据，并为以下 10 秒的视频间隔生成嵌入：[10, 20), [20, 30), [30, 40), [40, 50), [50, 60)。 此视频间隔 ("intervalSec": 10) 属于标准视频嵌入模式，用户按标准模式价格费率计费。

    如果省略 videoSegmentConfig，则服务使用以下默认值： "videoSegmentConfig": { "startOffsetSec": 0, "endOffsetSec": 120, "intervalSec": 16 }。 此视频间隔 ("intervalSec": 16) 属于基本视频嵌入模式，用户按基本模式价格费率计费。

HTTP 方法和网址：

POST https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict

请求 JSON 正文： 

{
  "instances": [
    {
      "text": "TEXT",
      "image": {
        "gcsUri": "IMAGE_URI"
      },
      "video": {
        "gcsUri": "VIDEO_URI",
        "videoSegmentConfig": {
          "startOffsetSec": START_SECOND,
          "endOffsetSec": END_SECOND,
          "intervalSec": INTERVAL_SECONDS
        }
      }
    }
  ]
}

如需发送请求，请选择以下方式之一：

curl

将请求正文保存在名为 request.json 的文件中，然后执行以下命令： 

curl -X POST \
     -H "Authorization: Bearer $(gcloud auth print-access-token)" \
     -H "Content-Type: application/json; charset=utf-8" \
     -d @request.json \
     "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict"

PowerShell

将请求正文保存在名为 request.json 的文件中，然后执行以下命令： 

$cred = gcloud auth print-access-token
$headers = @{ "Authorization" = "Bearer $cred" }

Invoke-WebRequest `
    -Method POST `
    -Headers $headers `
    -ContentType: "application/json; charset=utf-8" `
    -InFile request.json `
    -Uri "https://LOCATION-aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/publishers/google/models/multimodalembedding@001:predict" | Select-Object -Expand Content
模型返回的嵌入是 1408 个浮点矢量。以下示例响应会缩短。 
{
  "predictions": [
    {
      "textEmbedding": [
        0.0105433334,
        -0.00302835181,
        0.00656806398,
        0.00603460241,
        [...]
        0.00445805816,
        0.0139605571,
        -0.00170318608,
        -0.00490092579
      ],
      "videoEmbeddings": [
        {
          "startOffsetSec": 0,
          "endOffsetSec": 7,
          "embedding": [
            -0.00673126569,
            0.0248149596,
            0.0128901172,
            0.0107588246,
            [...]
            -0.00180952181,
            -0.0054573305,
            0.0117037306,
            0.0169312079
          ]
        }
      ],
      "imageEmbedding": [
        -0.00728622358,
        0.031021487,
        -0.00206603738,
        0.0273937676,
        [...]
        -0.00204976718,
        0.00321615417,
        0.0121978866,
        0.0193375275
      ]
    }
  ],
  "deployedModelId": "DEPLOYED_MODEL_ID"
}

Python

如需了解如何安装或更新 Vertex AI SDK for Python，请参阅安装 Vertex AI SDK for Python。 如需了解详情，请参阅 Python API 参考文档。 

import vertexai

from vertexai.vision_models import Image, MultiModalEmbeddingModel, Video
from vertexai.vision_models import VideoSegmentConfig

# TODO(developer): Update & uncomment line below
# PROJECT_ID = "your-project-id"
vertexai.init(project=PROJECT_ID, location="us-central1")

model = MultiModalEmbeddingModel.from_pretrained("multimodalembedding@001")

image = Image.load_from_file(
    "gs://cloud-samples-data/vertex-ai/llm/prompts/landmark1.png"
)
video = Video.load_from_file(
    "gs://cloud-samples-data/vertex-ai-vision/highway_vehicles.mp4"
)

embeddings = model.get_embeddings(
    image=image,
    video=video,
    video_segment_config=VideoSegmentConfig(end_offset_sec=1),
    contextual_text="Cars on Highway",
)

print(f"Image Embedding: {embeddings.image_embedding}")

# Video Embeddings are segmented based on the video_segment_config.
print("Video Embeddings:")
for video_embedding in embeddings.video_embeddings:
    print(
        f"Video Segment: {video_embedding.start_offset_sec} - {video_embedding.end_offset_sec}"
    )
    print(f"Embedding: {video_embedding.embedding}")

print(f"Text Embedding: {embeddings.text_embedding}")
# Example response:
# Image Embedding: [-0.0123144267, 0.0727186054, 0.000201397663, ...]
# Video Embeddings:
# Video Segment: 0.0 - 1.0
# Embedding: [-0.0206376351, 0.0345234685, ...]
# Text Embedding: [-0.0207006838, -0.00251058186, ...]

Go

在尝试此示例之前，请按照《Vertex AI 快速入门：使用客户端库》中的 Go 设置说明执行操作。 如需了解详情，请参阅 Vertex AI Go API 参考文档。

如需向 Vertex AI 进行身份验证，请设置应用默认凭证。 如需了解详情，请参阅为本地开发环境设置身份验证。 

import (
	"context"
	"encoding/json"
	"fmt"
	"io"
	"time"

	aiplatform "cloud.google.com/go/aiplatform/apiv1beta1"
	aiplatformpb "cloud.google.com/go/aiplatform/apiv1beta1/aiplatformpb"
	"google.golang.org/api/option"
	"google.golang.org/protobuf/encoding/protojson"
	"google.golang.org/protobuf/types/known/structpb"
)

// generateForImageTextAndVideo shows how to use the multimodal model to generate embeddings for
// image, text and video data.
func generateForImageTextAndVideo(w io.Writer, project, location string) error {
	// location = "us-central1"

	// The default context timeout may be not enough to process a video input.
	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
	defer cancel()

	apiEndpoint := fmt.Sprintf("%s-aiplatform.googleapis.com:443", location)
	client, err := aiplatform.NewPredictionClient(ctx, option.WithEndpoint(apiEndpoint))
	if err != nil {
		return fmt.Errorf("failed to construct API client: %w", err)
	}
	defer client.Close()

	model := "multimodalembedding@001"
	endpoint := fmt.Sprintf("projects/%s/locations/%s/publishers/google/models/%s", project, location, model)

	// This is the input to the model's prediction call. For schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#request_body
	instance, err := structpb.NewValue(map[string]any{
		"text": "Domestic cats in natural conditions",
		"image": map[string]any{
			// Image and video inputs can be provided either as a Google Cloud Storage URI or as
			// base64-encoded bytes using the "bytesBase64Encoded" field.
			"gcsUri": "gs://cloud-samples-data/generative-ai/image/320px-Felis_catus-cat_on_snow.jpg",
		},
		"video": map[string]any{
			"gcsUri": "gs://cloud-samples-data/video/cat.mp4",
		},
	})
	if err != nil {
		return fmt.Errorf("failed to construct request payload: %w", err)
	}

	req := &aiplatformpb.PredictRequest{
		Endpoint: endpoint,
		// The model supports only 1 instance per request.
		Instances: []*structpb.Value{instance},
	}

	resp, err := client.Predict(ctx, req)
	if err != nil {
		return fmt.Errorf("failed to generate embeddings: %w", err)
	}

	instanceEmbeddingsJson, err := protojson.Marshal(resp.GetPredictions()[0])
	if err != nil {
		return fmt.Errorf("failed to convert protobuf value to JSON: %w", err)
	}
	// For response schema, see:
	// https://cloud.google.com/vertex-ai/generative-ai/docs/model-reference/multimodal-embeddings-api#response-body
	var instanceEmbeddings struct {
		ImageEmbeddings []float32 `json:"imageEmbedding"`
		TextEmbeddings  []float32 `json:"textEmbedding"`
		VideoEmbeddings []struct {
			Embedding      []float32 `json:"embedding"`
			StartOffsetSec float64   `json:"startOffsetSec"`
			EndOffsetSec   float64   `json:"endOffsetSec"`
		} `json:"videoEmbeddings"`
	}
	if err := json.Unmarshal(instanceEmbeddingsJson, &instanceEmbeddings); err != nil {
		return fmt.Errorf("failed to unmarshal JSON: %w", err)
	}

	imageEmbedding := instanceEmbeddings.ImageEmbeddings
	textEmbedding := instanceEmbeddings.TextEmbeddings
	// Get the embedding for our single video segment (`.videoEmbeddings` object has one entry per
	// each processed segment).
	videoEmbedding := instanceEmbeddings.VideoEmbeddings[0].Embedding

	fmt.Fprintf(w, "Image embedding (length=%d): %v\n", len(imageEmbedding), imageEmbedding)
	fmt.Fprintf(w, "Text embedding (length=%d): %v\n", len(textEmbedding), textEmbedding)
	fmt.Fprintf(w, "Video embedding (length=%d): %v\n", len(videoEmbedding), videoEmbedding)
	// Example response:
	// Image embedding (length=1408): [-0.01558477 0.0258355 0.016342038 ... ]
	// Text embedding (length=1408): [-0.005894961 0.008349559 0.015355394 ... ]
	// Video embedding (length=1408): [-0.018867437 0.013997682 0.0012682161 ... ]

	return nil
}

后续步骤

如需查看详细文档，请参阅以下内容：