app.py

import os
import requests
from tqdm import tqdm
from datasets import load_dataset
import numpy as np
from tensorflow.keras.applications.resnet50 import ResNet50, preprocess_input
from tensorflow.keras.preprocessing import image
from sklearn.neighbors import NearestNeighbors
import joblib
from PIL import UnidentifiedImageError, Image
import gradio as gr
import matplotlib.pyplot as plt

# Load the dataset
dataset = load_dataset("thefcraft/civitai-stable-diffusion-337k")

# Take a subset of the dataset
subset_size = 50
dataset_subset = dataset['train'].shuffle(seed=42).select(range(subset_size))

# Directory to save images
image_dir = 'civitai_images'
os.makedirs(image_dir, exist_ok=True)

# Load the ResNet50 model pretrained on ImageNet
model = ResNet50(weights='imagenet', include_top=False, pooling='avg')

# Function to extract features
def extract_features(img_path, model):
    img = image.load_img(img_path, target_size=(224, 224))
    img_array = image.img_to_array(img)
    img_array = np.expand_dims(img_array, axis=0)
    img_array = preprocess_input(img_array)
    features = model.predict(img_array)
    return features.flatten()

# Extract features for a sample of images
features = []
image_paths = []
model_names = []

for sample in tqdm(dataset_subset):
    img_url = sample['url']  # Adjust based on the correct column name
    model_name = sample['Model']  # Adjust based on the correct column name
    img_path = os.path.join(image_dir, os.path.basename(img_url))

    # Download the image
    try:
        response = requests.get(img_url)
        response.raise_for_status()  # Check if the download was successful

        if 'image' not in response.headers['Content-Type']:
            raise ValueError("URL does not contain an image")

        with open(img_path, 'wb') as f:
            f.write(response.content)

        # Extract features
        try:
            img_features = extract_features(img_path, model)
            features.append(img_features)
            image_paths.append(img_path)
            model_names.append(model_name)
        except UnidentifiedImageError:
            print(f"UnidentifiedImageError: Skipping file {img_path}")
            os.remove(img_path)

    except requests.exceptions.RequestException as e:
        print(f"RequestException: Failed to download {img_url} - {e}")

# Convert features to numpy array
features = np.array(features)

# Build the NearestNeighbors model
nbrs = NearestNeighbors(n_neighbors=5, algorithm='ball_tree').fit(features)

# Save the model and features
joblib.dump(nbrs, 'nearest_neighbors_model.pkl')
np.save('image_features.npy', features)
np.save('image_paths.npy', image_paths)
np.save('model_names.npy', model_names)

# Load the NearestNeighbors model and features
nbrs = joblib.load('nearest_neighbors_model.pkl')
features = np.load('image_features.npy')
image_paths = np.load('image_paths.npy', allow_pickle=True)
model_names = np.load('model_names.npy', allow_pickle=True)

# Function to get recommendations
def get_recommendations(img_path, model, nbrs, image_paths, model_names, n_neighbors=5):
    img_features = extract_features(img_path, model)
    distances, indices = nbrs.kneighbors([img_features])

    recommended_images = [image_paths[idx] for idx in indices.flatten()]
    recommended_model_names = [model_names[idx] for idx in indices.flatten()]
    recommended_distances = distances.flatten()

    return recommended_images, recommended_model_names, recommended_distances

def recommend(image):
    # Save uploaded image to a path
    image_path = "uploaded_image.jpg"
    image.save(image_path)
    
    recommended_images, recommended_model_names, recommended_distances = get_recommendations(image_path, model, nbrs, image_paths, model_names)
    result = list(zip(recommended_images, recommended_model_names, recommended_distances))
    
    # Prepare HTML output for Gradio
    html_output = ""
    for img_path, model_name, distance in zip(recommended_images, recommended_model_names, recommended_distances):
        img_path = img_path.replace('\\', '/')
        html_output += f"""
        <div style='display:inline-block; text-align:center; margin:10px;'>
            <img src='file/{img_path}' style='width:200px; height:200px;'><br>
            <b>Model Name:</b> {model_name}<br>
            <b>Distance:</b> {distance:.2f}<br>
        </div>
        """
    
    return html_output

interface = gr.Interface(
    fn=recommend,
    inputs=gr.Image(type="pil"),
    outputs=gr.HTML(),  # Use HTML output for better formatting
    title="Image Recommendation System",
    description="Upload an image and get 5 recommended similar images with model names and distances."
)

interface.launch()