Visual Image Similarity Search

import json, pathlib, tarfile, urllib.request
from PIL import Image
from sentence_transformers import SentenceTransformer
from veep import Client

urllib.request.urlretrieve(
    "https://next.vectorpanda.com/data/example-met-famous.tar.gz",
    "met-famous.tar.gz",
)
with tarfile.open("met-famous.tar.gz") as tf:
    tf.extractall(".")

model = SentenceTransformer("clip-ViT-B-32")   # text and images share one space
vp = Client.login()

meta = json.loads(pathlib.Path("met-famous/metadata.json").read_text())
vp.collections.create("artworks", tier="hot")
vp.vectors.upsert("artworks", vectors=[
    {"id": str(row["object_id"]),
     "vector": model.encode(Image.open(f"met-famous/{row['image_file']}")).tolist(),
     "metadata": {"title": row["title"], "artist": row["artist"]}}
    for row in meta
])

q = model.encode("a swirling night sky over a quiet village").tolist()
for r in vp.vectors.query("artworks", q, top_k=3):
    print(f"{r.metadata['title']} — {r.metadata['artist']}  score={r.score:.3f}")