import numpy as np
from PIL import Image
import torch
from dotenv import load_dotenv

# Load environment variables
load_dotenv()

def get_image_embedding(image_path):
    """
    Get image embedding using a local CLIP model.

    Args:
        image_path: Path to the image file

    Returns:
        Embedding as numpy array, or None if embedding failed
    """
    try:
        # Import here to avoid immediate dependency requirements
        import clip

        # Load CLIP model
        device = "cuda" if torch.cuda.is_available() else "cpu"
        model, preprocess = clip.load("ViT-B/32", device=device)

        # Load and preprocess the image
        image = preprocess(Image.open(image_path)).unsqueeze(0).to(device)

        # Generate the embedding
        with torch.no_grad():
            image_features = model.encode_image(image)

        # Convert to numpy and normalize
        embedding = image_features.cpu().numpy()[0]
        embedding = embedding / np.linalg.norm(embedding)

        return embedding

    except ImportError:
        print("CLIP not installed. Falling back to simpler embedding method.")
        return get_simple_embedding(image_path)
    except Exception as e:
        print(f"[CLIP] Error: {e}")
        return None

def get_simple_embedding(image_path):
    """
    A simpler fallback embedding method using image histograms.
    This is much less sophisticated than CLIP but can work as a backup.

    Args:
        image_path: Path to the image file

    Returns:
        Embedding as numpy array
    """
    try:
        # Open image and convert to grayscale
        image = Image.open(image_path).convert('L')

        # Resize for consistency
        image = image.resize((224, 224))

        # Convert to numpy array
        img_array = np.array(image)

        # Create a simple histogram-based feature
        hist, _ = np.histogram(img_array, bins=64, range=(0, 256))

        # Normalize the histogram
        embedding = hist / np.linalg.norm(hist)

        return embedding
    except Exception as e:
        print(f"[Simple Embedding] Error: {e}")
        return None