import torch
import torchvision.transforms as transforms
from PIL import Image
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from torchvision.models import resnet50
from torchvision.datasets import ImageFolder
from torch.utils.data import DataLoader

# Load a pre-trained ResNet-50 model
model = resnet50(pretrained=True)
model.eval()

# Define a function to preprocess images
def preprocess_image(image_path):
    transform = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])
    image = Image.open(image_path)
    image = transform(image).unsqueeze(0)  # Add a batch dimension
    return image

# Load your ideal subset of images
ideal_image_paths = ["/content/trunck.jpg", "t4.jpg"]  # Replace with your ideal image file paths
ideal_embeddings = []

for image_path in ideal_image_paths:
    image = preprocess_image(image_path)
    with torch.no_grad():
        embedding = model(image).squeeze().numpy()
        ideal_embeddings.append(embedding)

# Load a set of candidate images
candidate_image_paths = ["/content/trunck2.jpg", "t3.jpg", "car.jpg",]  # Replace with your candidate image file paths
candidate_embeddings = []

for image_path in candidate_image_paths:
    image = preprocess_image(image_path)
    with torch.no_grad():
        embedding = model(image).squeeze().numpy()
        candidate_embeddings.append(embedding)

# Calculate similarities between ideal and candidate images using cosine similarity
similarities = cosine_similarity(ideal_embeddings, candidate_embeddings)

# Print the similarity matrix
print(similarities)


## SWIN code

import torch
from transformers import SwinTransformer, SwinTransformerImageProcessor
import torchvision.transforms as transforms
from PIL import Image
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity

# Load the pretrained Swin Transformer model and image processor
model_name = "microsoft/Swin-Transformer-base-patch4-in22k"
model = SwinTransformer.from_pretrained(model_name)
processor = SwinTransformerImageProcessor.from_pretrained(model_name)

# Define a function to preprocess images
def preprocess_image(image_path):
    transform = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
    ])
    image = Image.open(image_path)
    inputs = processor(images=image, return_tensors="pt")
    return inputs

# Load your ideal and candidate subsets of images
ideal_image_paths = ["ideal_image1.jpg", "ideal_image2.jpg", "ideal_image3.jpg"]  # Replace with your ideal image file paths
candidate_image_paths = ["candidate_image1.jpg", "candidate_image2.jpg", "candidate_image3.jpg"]  # Replace with your candidate image file paths

# Calculate embeddings for ideal images
ideal_embeddings = []
for image_path in ideal_image_paths:
    inputs = preprocess_image(image_path)
    with torch.no_grad():
        output = model(**inputs)
        embedding = output['pixel_values'][0].cpu().numpy()
        ideal_embeddings.append(embedding)

# Calculate embeddings for candidate images
candidate_embeddings = []
for image_path in candidate_image_paths:
    inputs = preprocess_image(image_path)
    with torch.no_grad():
        output = model(**inputs)
        embedding = output['pixel_values'][0].cpu().numpy()
        candidate_embeddings.append(embedding)

# Calculate cosine similarities between ideal and candidate images
similarities = cosine_similarity(ideal_embeddings, candidate_embeddings)

# Set a similarity threshold (e.g., 0.7)
threshold = 0.7

# Find similar image pairs based on the threshold
similar_pairs = []
for i in range(len(ideal_image_paths)):
    for j in range(len(candidate_image_paths)):
        if similarities[i, j] > threshold:
            similar_pairs.append((ideal_image_paths[i], candidate_image_paths[j]))

# Print similar image pairs
for pair in similar_pairs:
    print(f"Similar images: {pair[0]} and {pair[1]}")