AI & Vectors

Semantic Image Search with Amazon Titan

Implement semantic image search with Amazon Titan and Supabase Vector in Python.

Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from leading AI companies like AI21 Labs, Anthropic, Cohere, Meta, Mistral AI, Stability AI, and Amazon. Each model is accessible through a common API which implements a broad set of features to help build generative AI applications with security, privacy, and responsible AI in mind.

Amazon Titan is a family of foundation models (FMs) for text and image generation, summarization, classification, open-ended Q&A, information extraction, and text or image search.

In this guide we'll look at how we can get started with Amazon Bedrock and Supabase Vector in Python using the Amazon Titan multimodal model and the vecs client.

You can find the full application code as a Python Poetry project on GitHub.

Create a new Python project with Poetry

Poetry provides packaging and dependency management for Python. If you haven't already, install poetry via pip:

1
pip install poetry

Then initialize a new project:

1
poetry new aws_bedrock_image_search

Spin up a Postgres database with pgvector

If you haven't already, head over to database.new and create a new project. Every Supabase project comes with a full Postgres database and the pgvector extension preconfigured.

When creating your project, make sure to note down your database password as you will need it to construct the DB_URL in the next step.

You can find the database connection string in your Supabase Dashboard database settings. Select "Use connection pooling" with Mode: Session for a direct connection to your Postgres database. It will look something like this:

1
postgresql://postgres.[PROJECT-REF]:[YOUR-PASSWORD]@aws-0-[REGION].pooler.supabase.com:5432/postgres

Install the dependencies

We will need to add the following dependencies to our project:

  • vecs: Supabase Vector Python Client.
  • boto3: AWS SDK for Python.
  • matplotlib: for displaying our image result.
1
poetry add vecs boto3 matplotlib

Import the necessary dependencies

At the top of your main python script, import the dependencies and store your DB URL from above in a variable:

1
2
3
4
5
6
7
8
9
10
import sysimport boto3import vecsimport jsonimport base64from matplotlib import pyplot as pltfrom matplotlib import image as mpimgfrom typing import OptionalDB_CONNECTION = "postgresql://postgres.[PROJECT-REF]:[YOUR-PASSWORD]@aws-0-[REGION].pooler.supabase.com:5432/postgres"

Next, get the credentials to your AWS account and instantiate the boto3 client:

1
2
3
4
5
6
7
8
bedrock_client = boto3.client(    'bedrock-runtime',    region_name='us-west-2',    # Credentials from your AWS account    aws_access_key_id='<replace_your_own_credentials>',    aws_secret_access_key='<replace_your_own_credentials>',    aws_session_token='<replace_your_own_credentials>',)

Create embeddings for your images

In the root of your project, create a new folder called images and add some images. You can use the images from the example project on GitHub or you can find license free images on Unsplash.

To send images to the Amazon Bedrock API we need to need to encode them as base64 strings. Create the following helper methods:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
def readFileAsBase64(file_path):    """Encode image as base64 string."""    try:        with open(file_path, "rb") as image_file:            input_image = base64.b64encode(image_file.read()).decode("utf8")        return input_image    except:        print("bad file name")        sys.exit(0)def construct_bedrock_image_body(base64_string):    """Construct the request body.    https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-titan-embed-mm.html    """    return json.dumps(        {            "inputImage": base64_string,            "embeddingConfig": {"outputEmbeddingLength": 1024},        }    )def get_embedding_from_titan_multimodal(body):    """Invoke the Amazon Titan Model via API request."""    response = bedrock_client.invoke_model(        body=body,        modelId="amazon.titan-embed-image-v1",        accept="application/json",        contentType="application/json",    )    response_body = json.loads(response.get("body").read())    print(response_body)    return response_body["embedding"]def encode_image(file_path):    """Generate embedding for the image at file_path."""    base64_string = readFileAsBase64(file_path)    body = construct_bedrock_image_body(base64_string)    emb = get_embedding_from_titan_multimodal(body)    return emb

Next, create a seed method, which will create a new Supabase Vector Collection, generate embeddings for your images, and upsert the embeddings into your database:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
def seed():    # create vector store client    vx = vecs.create_client(DB_CONNECTION)    # get or create a collection of vectors with 1024 dimensions    images = vx.get_or_create_collection(name="image_vectors", dimension=1024)    # Generate image embeddings with Amazon Titan Model    img_emb1 = encode_image('./images/one.jpg')    img_emb2 = encode_image('./images/two.jpg')    img_emb3 = encode_image('./images/three.jpg')    img_emb4 = encode_image('./images/four.jpg')    # add records to the *images* collection    images.upsert(        records=[            (                "one.jpg",       # the vector's identifier                img_emb1,        # the vector. list or np.array                {"type": "jpg"}  # associated  metadata            ), (                "two.jpg",                img_emb2,                {"type": "jpg"}            ), (                "three.jpg",                img_emb3,                {"type": "jpg"}            ), (                "four.jpg",                img_emb4,                {"type": "jpg"}            )        ]    )    print("Inserted images")    # index the collection for fast search performance    images.create_index()    print("Created index")

Add this method as a script in your pyproject.toml file:

1
2
3
[tool.poetry.scripts]seed = "image_search.main:seed"search = "image_search.main:search"

After activating the virtual environment with poetry shell you can now run your seed script via poetry run seed. You can inspect the generated embeddings in your Supabase Dashboard by visiting the Table Editor, selecting the vecs schema, and the image_vectors table.

Perform an image search from a text query

We can use Supabase Vector to query our embeddings. We can either use an image as the search input or generate an embedding from a string input:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
def search(query_term: Optional[str] = None):    if query_term is None:        query_term = sys.argv[1]    # create vector store client    vx = vecs.create_client(DB_CONNECTION)    images = vx.get_or_create_collection(name="image_vectors", dimension=1024)    # Encode text query    text_emb = get_embedding_from_titan_multimodal(json.dumps(        {            "inputText": query_term,            "embeddingConfig": {"outputEmbeddingLength": 1024},        }    ))    # query the collection filtering metadata for "type" = "jpg"    results = images.query(        data=text_emb,                      # required        limit=1,                            # number of records to return        filters={"type": {"$eq": "jpg"}},   # metadata filters    )    result = results[0]    print(result)    plt.title(result)    image = mpimg.imread('./images/' + result)    plt.imshow(image)    plt.show()

By limiting the query to one result, we can show the most relevant image to the user. Finally we use matplotlib to show the image result to the user.

Go ahead and test it out by running poetry run search and you will be presented with an image of a "bike in front of a red brick wall".

Conclusion

With just a couple of lines of Python you are able to implement image search as well as reverse image search using the Amazon Titan multimodal model and Supabase Vector.