search.py

from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np

# Use a compatible CLIP model
model = SentenceTransformer("clip-ViT-B-32")



def search_images_by_text(text, df, embeddings, top_k=30):
    text_emb = model.encode([text])
    filtered_embeddings = embeddings[df.index]
    sims = cosine_similarity(text_emb, filtered_embeddings)[0]
    top_indices = np.argsort(sims)[::-1][:top_k]
    return df.iloc[top_indices]

def get_similar_images(df, image_id, embeddings, top_k=5):
    index = int(image_id)  # adjust based on your ID setup
    query_emb = embeddings[index]
    sims = cosine_similarity([query_emb], embeddings)[0]
    top_indices = np.argsort(sims)[::-1][1:top_k+1]
    return df.iloc[top_indices]


from PIL import Image
import torchvision.transforms as T



# Make sure to use the same embedding method for your dataset images
def preprocess_image(img):
    # You may want to adapt this if you used a specific transform for your dataset
    transform = T.Compose([
        T.Resize((224, 224)),
        T.ToTensor(),
        T.Normalize(mean=(0.48145466, 0.4578275, 0.40821073),
                    std=(0.26862954, 0.26130258, 0.27577711))
    ])
    return transform(img).unsqueeze(0)

def search_images_by_image(uploaded_image, df, embeddings, top_k=30):
    # Convert to PIL Image if necessary
    if isinstance(uploaded_image, str):  # if path
        uploaded_image = Image.open(uploaded_image).convert("RGB")
    elif isinstance(uploaded_image, np.ndarray):
        uploaded_image = Image.fromarray(uploaded_image).convert("RGB")

    # Encode image using CLIP model
    image_emb = model.encode(uploaded_image)

    # Compute cosine similarity to dataset embeddings
    filtered_embeddings = embeddings[df.index]
    sims = cosine_similarity([image_emb], filtered_embeddings)[0]
    top_indices = np.argsort(sims)[::-1][:top_k]
    
    return df.iloc[top_indices]