Image Embedding

import open_clip
import torch
from PIL import Image

# Load model and preprocessing
model, _, preprocess = open_clip.create_model_and_transforms(
    'ViT-L-14', pretrained='laion2b_s32b_b82k'
)
model.eval()
tokenizer = open_clip.get_tokenizer('ViT-L-14')

# Embed an image
image = preprocess(Image.open('photo.jpg')).unsqueeze(0)
with torch.no_grad():
    image_features = model.encode_image(image)
    image_features /= image_features.norm(dim=-1, keepdim=True)

# Embed a text query
text = tokenizer(['a photo of a cat', 'a photo of a dog'])
with torch.no_grad():
    text_features = model.encode_text(text)
    text_features /= text_features.norm(dim=-1, keepdim=True)

# Compute similarity
similarity = (100.0 * image_features @ text_features.T).softmax(dim=-1)
print(f'Similarity: {similarity}')

from transformers import AutoProcessor, AutoModel
from PIL import Image
import torch

# Load SigLIP
processor = AutoProcessor.from_pretrained('google/siglip-so400m-patch14-384')
model = AutoModel.from_pretrained('google/siglip-so400m-patch14-384')

# Prepare inputs
image = Image.open('photo.jpg')
texts = ['a cat sleeping', 'a dog running', 'a sunset']

inputs = processor(
    text=texts,
    images=image,
    padding='max_length',
    return_tensors='pt'
)

# Get embeddings
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits_per_image  # image-text similarity
    probs = torch.sigmoid(logits)  # SigLIP uses sigmoid, not softmax

for text, prob in zip(texts, probs[0]):
    print(f'{text}: {prob:.3f}')

import open_clip
import torch
import faiss
import numpy as np
from PIL import Image
from pathlib import Path
from tqdm import tqdm

# Setup
model, _, preprocess = open_clip.create_model_and_transforms(
    'ViT-L-14', pretrained='laion2b_s32b_b82k'
)
model.eval()
tokenizer = open_clip.get_tokenizer('ViT-L-14')

# Index all images in a folder
image_paths = list(Path('photos/').glob('*.jpg'))
embeddings = []

for path in tqdm(image_paths, desc='Indexing'):
    image = preprocess(Image.open(path)).unsqueeze(0)
    with torch.no_grad():
        feat = model.encode_image(image)
        feat /= feat.norm(dim=-1, keepdim=True)
    embeddings.append(feat.numpy())

# Build FAISS index
embeddings = np.vstack(embeddings).astype('float32')
index = faiss.IndexFlatIP(embeddings.shape[1])  # Inner product = cosine sim
index.add(embeddings)

# Search with text
query = 'sunset at the beach'
text_tokens = tokenizer([query])
with torch.no_grad():
    query_feat = model.encode_text(text_tokens)
    query_feat /= query_feat.norm(dim=-1, keepdim=True)

D, I = index.search(query_feat.numpy().astype('float32'), k=5)
print('Top 5 matches:')
for i, (dist, idx) in enumerate(zip(D[0], I[0])):
    print(f'  {i+1}. {image_paths[idx].name} (score: {dist:.3f})')