models/embedding.py

import torch
from transformers import CLIPProcessor, CLIPModel
from config.config import Config
from PIL import Image

class ImageEmbedder:
    _instance = None

    def __new__(cls, model_name=Config.CLIP_MODEL_NAME):
        if not cls._instance:
            cls._instance = super().__new__(cls)
            cls._instance.model = CLIPModel.from_pretrained(model_name)
            cls._instance.processor = CLIPProcessor.from_pretrained(model_name)
        return cls._instance
    
    def generate_embedding(self, image_path: str) -> torch.Tensor:
        """Generate embedding for an image using CLIP."""
        image = Image.open(image_path).convert('RGB')
        inputs = self.processor(images=image, return_tensors="pt")
        
        with torch.no_grad():
            image_embeddings = self.model.get_image_features(**inputs)
        
        return image_embeddings.squeeze().numpy()