Multimodal vector search - images

APPLY = False

from superduper import superduper

db = superduper('mongomock:///test_db')

Get useful sample data

def getter():
    import subprocess
    subprocess.run([
        'curl', '-O', 'https://superduperdb-public-demo.s3.amazonaws.com/images_classification.zip',
    ])
    subprocess.run(['rm', '-rf', 'images'])
    subprocess.run(['rm', '-rf', '__MACOSX'])
    subprocess.run(['unzip', 'images_classification.zip'])
    subprocess.run(['rm', 'images_classification.zip'])
    import json
    from PIL import Image
    with open('images/images.json', 'r') as f:
        data = json.load(f)
    data = data[:100]
    data = [{'img': Image.open(r['image_path'])} for r in data]
    subprocess.run(['rm', '-rf', '__MACOSX'])
    subprocess.run(['rm', '-rf', 'images'])
    return data

if APPLY:
    data = getter()

Build multimodal embedding models

We define the output data type of a model as a vector for vector transformation.

from superduper.components.vector_index import sqlvector

output_datatype = sqlvector(shape=(1024,))

Then define two models, one for text embedding and one for image embedding.

import clip
from superduper import vector, imported
from superduper_torch import TorchModel

rn50 = imported(clip.load)('RN50', device='cpu')

compatible_model = TorchModel(
    identifier='clip_text',
    object=rn50[0],
    preprocess=lambda x: clip.tokenize(x)[0],
    postprocess=lambda x: x.tolist(),
    datatype=output_datatype,
    forward_method='encode_text',
)

embedding_model = TorchModel(
    identifier='clip_image',
    object=rn50[0].visual,
    preprocess=rn50[1],
    postprocess=lambda x: x.tolist(),
    datatype=output_datatype,
)

Because we use multimodal models, we define different keys to specify which model to use for embedding calculations in the vector_index.

indexing_key = 'img'
compatible_key = 'text'

Create vector-index

vector_index_name = 'my-vector-index'

from superduper import VectorIndex, Listener

vector_index = VectorIndex(
    vector_index_name,
    indexing_listener=Listener(
        key=indexing_key,
        select=db['docs'].select(),
        model=embedding_model,
        identifier='indexing-listener',
    ),
    compatible_listener=Listener(
        key=compatible_key,
        model=compatible_model,
        select=None,
        identifier='compatible-listener',
    )
)

from superduper import Application

application = Application(
    'image-vector-search',
    components=[vector_index],
)

if APPLY:
    db.apply(application, force=True)

Add the data

The order in which data is added is not important. However if your data requires a custom Schema in order to work, it's easier to add the Application first, and the data later. The advantage of this flexibility, is that once the Application is installed, it's waiting for incoming data, so that the Application is always up-to-date. This comes in particular handy with AI scenarios which need to respond to changing news.

if APPLY:
    from superduper import Document
    
    table_or_collection = db['docs']
    
    ids = db.execute(table_or_collection.insert([Document(r) for r in data]))

Perform a vector search

We can perform the vector searches using two types of data:

Text: By text description, we can find images similar to the text description.
Image: By using an image, we can find images similar to the provided image.

if APPLY:
    item = Document({compatible_key: "Find a black dog."})

if APPLY:
    from IPython.display import display
    search_image = data[0]
    display(search_image)
    item = Document(search_image)

Once we have this search target, we can execute a search as follows.

if APPLY:
    select = db['docs'].like(item, vector_index=vector_index_name, n=5).select()

    results = list(db.execute(select))

    from IPython.display import display
    for result in results:
        display(result[indexing_key])

Create a `Template`

from superduper import Template, Table, Schema
from superduper.components.dataset import RemoteData
from superduper_pillow import pil_image


template = Template(
    'multimodal_image_search',
    template=application,
    default_table=Table(
        'sample_multimodal_image_search', 
        schema=Schema(
            'sample_multimodal_image_search/schema',
            fields={'img': pil_image},
        ),
        data=RemoteData('sample_images', getter=getter),
    ),
    substitutions={'docs': 'table_name', 'cpu': 'device'},
    types={
        'device': {
            'type': 'str',
            'default': 'cpu',
        },
        'table_name': {
            'type': 'str',
            'default': 'sample_multimodal_image_search',
        },
    }
)

template.export('.')

template.template

vector_index.indexing_listener.select

Multimodal vector search - images

Get useful sample data​

Build multimodal embedding models​

Create vector-index​

Add the data​

Perform a vector search​

Create a Template​