import streamlit as st
import cv2
import numpy as np
import joblib
import os
import warnings
from tensorflow.keras.models import load_model

# Gereksiz sistem uyarılarını kapat / Silence system warnings
warnings.filterwarnings('ignore')
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3' 

# Sayfa Ayarları / Page Configuration
st.set_page_config(page_title="Pencil Sketch & Face Recognition", layout="wide")

# --- 1. MODELLERİ ÖNBELLEĞE ALARAK YÜKLE / LOAD MODELS WITH CACHING ---
@st.cache_resource
def load_all_models():
    m1_path = 'model1_lbph.yml'
    m2_path = 'model2_knn.pkl'
    m3_path = 'model3_cnn.keras'
    
    try:
        # LBPH
        m1 = cv2.face.LBPHFaceRecognizer_create()
        m1.read(m1_path)
        # KNN (Version ignore)
        m2 = joblib.load(m2_path)
        # CNN (No compile to avoid optimizer errors)
        m3 = load_model(m3_path, compile=False)
        return m1, m2, m3
    except Exception as e:
        return None, None, None

# Modelleri yükle
model1, model2, model3 = load_all_models()

# --- 2. SOL PANEL (SIDEBAR) / SIDEBAR DESIGN ---
st.sidebar.title("⚙️ Sistem Paneli / System Panel")
st.sidebar.divider()

if model1 is not None:
    st.sidebar.success("""
    ### ✅ Durum / Status:
    **Modeller Yüklendi! / Models Loaded!**
    
    * **LBPH:** Aktif / Active
    * **KNN:** Aktif / Active
    * **CNN:** Aktif / Active
    """)
    st.sidebar.info("""
    **ℹ️ Not / Note:**
    Resim yüklendiğinde 3 model aynı anda çalışır.
    (3 models will process simultaneously.)
    """)
else:
    st.sidebar.error("""
    ### ❌ Hata / Error:
    **Modeller Bulunamadı! / Models Not Found!**
    Lütfen dosyaları kontrol edin.
    """)

st.sidebar.divider()
st.sidebar.caption("🚀 Final Project - Computer Vision")

# --- 3. ANA ARAYÜZ / MAIN INTERFACE ---
st.title("🎨 Karakalem & Yüz Tanıma | Pencil Sketch & Face Recognition")
st.write("---")

# Karakalem Fonksiyonu / Sketch Function
def get_sketch(image):
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    blur = cv2.GaussianBlur(gray, (5,5), 0)
    edges = cv2.Canny(blur, 10, 70)
    ret, mask = cv2.threshold(edges, 250, 255, cv2.THRESH_BINARY_INV)
    return mask

# Dosya Yükleme / File Upload
uploaded_file = st.file_uploader("Bir Resim Seçin / Choose an Image", type=["jpg", "png", "jpeg"])

if uploaded_file is not None:
    # Görüntüyü oku / Read image
    file_bytes = np.frombuffer(uploaded_file.read(), np.uint8)
    img = cv2.imdecode(file_bytes, cv2.IMREAD_COLOR)
    
    if img is not None:
        # Görsel Sonuçlar / Visual Results
        col1, col2 = st.columns(2)
        
        with col1:
            st.subheader("🖼️ Orijinal / Original")
            # use_container_width yerine yeni standart olan width='stretch' kullanıldı
            st.image(cv2.cvtColor(img, cv2.COLOR_BGR2RGB), width=500)
            
        with col2:
            st.subheader("✍️ Karakalem / Sketch")
            sketch_res = get_sketch(img)
            st.image(sketch_res, width=500)

        # Tahmin Bölümü / Prediction Section
        if model1 is not None:
            st.divider()
            st.header("🤖 Model Analizleri / Model Analysis")
            t1, t2, t3 = st.columns(3)
            
            # LBPH Tahmin
            with t1:
                try:
                    gray_lb = cv2.resize(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), (200,200))
                    label, conf = model1.predict(gray_lb)
                    st.metric("LBPH Sonucu", f"ID: {label}", f"Güven: {round(conf,1)}")
                except: st.error("LBPH Error")

            # KNN Tahmin
            with t2:
                try:
                    gray_knn = cv2.resize(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), (100,100)).flatten().reshape(1,-1)
                    res2 = model2.predict(gray_knn)
                    st.metric("KNN Sonucu", str(res2[0]))
                except: st.error("KNN Error")

            # CNN Tahmin
            with t3:
                try:
                    gray_cnn = cv2.resize(cv2.cvtColor(img, cv2.COLOR_BGR2GRAY), (64,64)) / 255.0
                    res3 = model3.predict(gray_cnn.reshape(1,64,64,1), verbose=0)
                    st.metric("CNN Sınıf", np.argmax(res3))
                except: st.error("CNN Error")