Visual Similarity Search

from mixpeek import Mixpeek

client = Mixpeek(api_key="YOUR_API_KEY", namespace="visual-assets")

# 1. Buckets for images and for videos, and a multimodal_extractor collection over each
images = client.buckets.create(
    bucket_name="image-assets",
    bucket_schema={
        "properties": {
            "image": {
                "type": "image",
            },
        },
    },
)
videos = client.buckets.create(
    bucket_name="video-assets",
    bucket_schema={
        "properties": {
            "video": {
                "type": "video",
            },
        },
    },
)
image_col = client.collections.create(
    collection_name="image_assets",
    source={"type": "bucket", "bucket_ids": [images["bucket_id"]]},
    feature_extractor={
        "feature_extractor_name": "multimodal_extractor",
        "version": "v1",
    },
)
video_col = client.collections.create(
    collection_name="video_assets",
    source={"type": "bucket", "bucket_ids": [videos["bucket_id"]]},
    feature_extractor={
        "feature_extractor_name": "multimodal_extractor",
        "version": "v1",
    },
)

# 2. Upload and process
client.buckets.upload(
    images["bucket_id"],
    blobs=[{"property": "image", "type": "image", "data": "s3://your-bucket/images/hero-shot.jpg"}],
)
client.buckets.upload(
    videos["bucket_id"],
    blobs=[{"property": "video", "type": "video", "data": "s3://your-bucket/videos/campaign-cut.mp4"}],
)
client.collections.trigger(image_col["collection_id"])
client.collections.trigger(video_col["collection_id"])

# 3. One image query across both collections, diversified with MMR
retriever = client.retrievers.create(
    retriever_name="visual-similarity",
    collection_identifiers=["image_assets", "video_assets"],
    input_schema={
        "image": {
            "type": "image",
            "required": True,
        },
    },
    stages=[
        {
            "stage_name": "search",
            "stage_id": "feature_search",
            "parameters": {
                "searches": [
                    {
                        "feature_uri": "mixpeek://multimodal_extractor@v1/vertex_multimodal_embedding",
                        "query": {
                            "input_mode": "content",
                            "value": "{{INPUT.image}}",
                        },
                        "top_k": 50,
                    },
                ],
                "final_top_k": 50,
            },
        },
        {
            "stage_name": "diversify",
            "stage_id": "mmr",
            "parameters": {
                "lambda": 0.7,
                "diversity_feature_uri": "mixpeek://multimodal_extractor@v1/vertex_multimodal_embedding",
                "top_k": 20,
            },
        },
    ],
)

# 4. Search
results = client.retrievers.execute(
    retriever["retriever_id"],
    inputs={
    "image": "https://example.com/reference.jpg",
},
)

for doc in results["documents"]:
    print(doc["document_id"], doc["score"])