app.py

import os
import gradio as gr
from PIL import Image
import torch
import numpy as np
from facenet_pytorch import MTCNN, InceptionResnetV1
from torch.nn.functional import cosine_similarity
import requests
from bs4 import BeautifulSoup
from io import BytesIO

# Initialize models
mtcnn = MTCNN(image_size=160, margin=20, keep_all=False)
resnet = InceptionResnetV1(pretrained='vggface2').eval()

# Directory for temporary images
DB_DIR = "scraped_images"
os.makedirs(DB_DIR, exist_ok=True)

# Scrape image URLs from Bing
def fetch_image_urls(query, max_images=3):
    headers = {"User-Agent": "Mozilla/5.0"}
    search_url = f"https://www.bing.com/images/search?q={query.replace(' ', '+')}"
    response = requests.get(search_url, headers=headers)
    soup = BeautifulSoup(response.text, 'html.parser')
    image_elements = soup.find_all('a', class_='iusc')
    urls = []
    for elem in image_elements[:max_images]:
        m = elem.get('m')
        if m:
            try:
                m_json = eval(m)
                urls.append(m_json['murl'])
            except:
                continue
    return urls

# Download and save images
def download_images(image_urls):
    for i, url in enumerate(image_urls):
        try:
            response = requests.get(url)
            img = Image.open(BytesIO(response.content)).convert('RGB')
            img.save(os.path.join(DB_DIR, f"scraped_{i}.jpg"))
        except:
            continue

# Load embeddings for downloaded images
def load_scraped_embeddings():
    embeddings = {}
    for file in os.listdir(DB_DIR):
        if file.lower().endswith(('.jpg', '.png')):
            img_path = os.path.join(DB_DIR, file)
            img = Image.open(img_path).convert("RGB")
            face = mtcnn(img)
            if face is not None:
                face = face.unsqueeze(0)
                emb = resnet(face)
                embeddings[file] = emb
    return embeddings

# Compare uploaded face with scraped images
def identify_person(uploaded_img, search_query):
    if uploaded_img is None:
        return "Please upload an image."

    # Step 1: Scrape and download images
    for f in os.listdir(DB_DIR):
        os.remove(os.path.join(DB_DIR, f))
    image_urls = fetch_image_urls(search_query, max_images=5)
    download_images(image_urls)

    # Step 2: Load scraped embeddings
    db_embeddings = load_scraped_embeddings()

    # Step 3: Get uploaded face embedding
    uploaded_face = mtcnn(uploaded_img.convert("RGB"))
    if uploaded_face is None:
        return "No face detected in the uploaded image."
    uploaded_embedding = resnet(uploaded_face.unsqueeze(0))

    # Step 4: Compare and find best match
    best_match = None
    best_score = -1
    for name, emb in db_embeddings.items():
        score = cosine_similarity(uploaded_embedding, emb).item()
        if score > best_score:
            best_score = score
            best_match = name

    if best_score > 0.7:
        return f"Best Match: {best_match}\nSimilarity: {best_score:.2f}"
    else:
        return f"No confident match found.\nBest candidate: {best_match} (Score: {best_score:.2f})"

# Gradio Interface
iface = gr.Interface(
    fn=identify_person,
    inputs=[gr.Image(type="pil"), gr.Text(label="Search Query (e.g., Elon Musk)")],
    outputs="text",
    title="Reverse Face Search via Web Scraping",
    description="Upload a face image and enter a name or query. The app scrapes the web for images and tries to identify the person."
)

if __name__ == "__main__":
    iface.launch()