app.py

import numpy as np
from PIL import Image, UnidentifiedImageError
import gradio as gr
from deepface import DeepFace
from datasets import load_dataset, Image as HfImage
import os
import pickle
from pathlib import Path
import gc
import requests
from io import BytesIO

# 🔑 Token de autenticación
HF_TOKEN = os.getenv("HF_TOKEN")
if not HF_TOKEN:
    raise ValueError("⚠️ Por favor, configura la variable de entorno HF_TOKEN para acceder al dataset privado")

# 📁 Directorio para embeddings
EMBEDDINGS_DIR = Path("embeddings")
EMBEDDINGS_DIR.mkdir(exist_ok=True)
EMBEDDINGS_FILE = EMBEDDINGS_DIR / "embeddings.pkl"

# ✅ Cargar dataset desde metadata.csv (con URLs absolutas)
dataset = load_dataset("csv", data_files="metadata.csv")

# 🔄 Preprocesar imagen para Facenet
def preprocess_image(img: Image.Image) -> np.ndarray:
    img_rgb = img.convert("RGB")
    img_resized = img_rgb.resize((160, 160), Image.Resampling.LANCZOS)
    return np.array(img_resized)

# 📦 Construir base de datos de embeddings
def build_database():
    if EMBEDDINGS_FILE.exists():
        print("📂 Cargando embeddings desde el archivo...")
        with open(EMBEDDINGS_FILE, 'rb') as f:
            return pickle.load(f)

    print("🔄 Calculando embeddings (esto puede tomar unos minutos)...")
    database = []
    batch_size = 10

    # Get the train split
    train_dataset = dataset["train"]
    
    # Debug: Print dataset structure
    print("Dataset structure:", train_dataset.features)
    print("First item structure:", train_dataset[0])
    print("Dataset type:", type(train_dataset))
    print("Dataset item type:", type(train_dataset[0]))

    for i in range(0, len(train_dataset), batch_size):
        batch = train_dataset[i:i + batch_size]
        print(f"📦 Procesando lote {i // batch_size + 1}/{(len(train_dataset) + batch_size - 1) // batch_size}")

        for j, item in enumerate(batch):
            try:
                print(f"Debug - Processing item {i+j}")
                print(f"Debug - Item type: {type(item)}")
                print(f"Debug - Item content: {item}")
                
                # Get the image URL
                image_url = item["image"]
                if not isinstance(image_url, str) or not image_url.startswith("http"):
                    print(f"⚠️ Skipping item {i+j} - Invalid URL format")
                    continue
                
                # Download and process the image
                response = requests.get(image_url, timeout=10)
                response.raise_for_status()
                img = Image.open(BytesIO(response.content))
                
                # Ensure image is in RGB mode
                img = img.convert("RGB")
                img_processed = preprocess_image(img)
                embedding = DeepFace.represent(
                    img_path=img_processed,
                    model_name="Facenet",
                    enforce_detection=False
                )[0]["embedding"]

                database.append((f"image_{i+j}", img, embedding))
                print(f"✅ Procesada imagen {i+j+1}/{len(train_dataset)}")

                del img_processed
                gc.collect()

            except Exception as e:
                print(f"❌ No se pudo procesar imagen {i+j}: {str(e)}")
                print(f"Error details: {type(e).__name__}")
                import traceback
                print(traceback.format_exc())
                continue

        # 💾 Guardar después de cada batch
        if database:
            print("💾 Guardando progreso...")
            with open(EMBEDDINGS_FILE, 'wb') as f:
                pickle.dump(database, f)

        gc.collect()

    return database

# 🔍 Buscar rostros similares
def find_similar_faces(uploaded_image: Image.Image):
    try:
        img_processed = preprocess_image(uploaded_image)
        query_embedding = DeepFace.represent(
            img_path=img_processed,
            model_name="Facenet",
            enforce_detection=False
        )[0]["embedding"]
        del img_processed
        gc.collect()
    except Exception as e:
        print(f"Error al procesar imagen de consulta: {str(e)}")
        return [], "⚠ No se detectó un rostro válido en la imagen."

    similarities = []
    for name, db_img, embedding in database:
        dist = np.linalg.norm(np.array(query_embedding) - np.array(embedding))
        sim_score = 1 / (1 + dist)
        similarities.append((sim_score, name, db_img))

    similarities.sort(reverse=True)
    top_matches = similarities[:5]

    gallery_items = []
    text_summary = ""
    for sim, name, img in top_matches:
        caption = f"{name} - Similitud: {sim:.2f}"
        gallery_items.append((img, caption))
        text_summary += caption + "\n"

    return gallery_items, text_summary

# ⚙️ Inicializar
print("🚀 Iniciando aplicación...")
database = build_database()
print(f"✅ Base de datos cargada con {len(database)} imágenes")

# 🎛️ Interfaz Gradio
demo = gr.Interface(
    fn=find_similar_faces,
    inputs=gr.Image(label="📤 Sube una imagen", type="pil"),
    outputs=[
        gr.Gallery(label="📸 Rostros más similares"),
        gr.Textbox(label="🧠 Similitud", lines=6)
    ],
    title="🔍 Buscador de Rostros con DeepFace",
    description="Sube una imagen y se comparará contra los rostros del dataset alojado en Hugging Face (`Segizu/facial-recognition`)."
)

demo.launch()