import cv2
import streamlit as st
import tempfile
import torch
import torch.nn as nn
from torchvision import transforms
from mtcnn import MTCNN
from skimage.feature import hog
import joblib
import numpy as np

class VGGFaceEmbedding(nn.Module):
    def __init__(self):
        super(VGGFaceEmbedding, self).__init__()
        self.base_model = resnet50(pretrained=True)
        self.base_model = nn.Sequential(*list(self.base_model.children())[:-2])
        self.pooling = nn.AdaptiveAvgPool2d((1, 1))
        self.flatten = nn.Flatten()

    def forward(self, x):
        x = self.base_model(x)
        x = self.pooling(x)
        x = self.flatten(x)
        return x

class L1Dist(nn.Module):
    def __init__(self):
        super(L1Dist, self).__init__()

    def forward(self, input_embedding, validation_embedding):
        return torch.abs(input_embedding - validation_embedding)

class SiameseNetwork(nn.Module):
    def __init__(self):
        super(SiameseNetwork, self).__init__()
        self.embedding = VGGFaceEmbedding()
        self.distance = L1Dist()
        self.fc1 = nn.Linear(2048, 512)
        self.fc2 = nn.Linear(512, 1)
        self.sigmoid = nn.Sigmoid()

    def forward(self, input_image, validation_image):
        input_embedding = self.embedding(input_image)
        validation_embedding = self.embedding(validation_image)
        distances = self.distance(input_embedding, validation_embedding)
        x = self.fc1(distances)
        x = self.fc2(x)
        x = self.sigmoid(x)
        return x

def preprocess_image_siamese(img):
    transform = transforms.Compose([
        transforms.Resize((224, 224)),
        transforms.ToTensor()
    ])
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    return transform(img)

def preprocess_image_svm(img):
    img = cv2.resize(img, (224, 224))
    img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    return img

def extract_hog_features(img):
    hog_features = hog(img, orientations=9, pixels_per_cell=(16, 16), cells_per_block=(4, 4))
    return hog_features

def get_face(img):
    detector = MTCNN()
    faces = detector.detect_faces(img)
    if faces:
        x1, y1, w, h = faces[0]['box']
        x1, y1 = abs(x1), abs(y1)
        x2, y2 = x1 + w, y1 + h
        return img[y1:y2, x1:x2]
    return None

def verify(image, model, person):
    
    with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_image:
        temp_image.write(image.read())
        temp_image_path = temp_image.name
        
    image = cv2.imread(temp_image_path)

    face = get_face(image)

    if face is not None:
        if model == "Siamese":
            siamese = SiameseNetwork()
            siamese.load_state_dict(torch.load(f'siamese_{person.lower()}.pth'))

            face = preprocess_image_siamese(face)

            # Move to device
            device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
            model.to(device)
            face = face.to(device)
        
            with torch.no_grad():
                output = model(face)
                probability = output.item()
                pred = 1.0 if probability > 0.5 else 0.0
        
            if pred == 1:
                st.write("Match")
            else:
                st.write("Not Match")
        
        elif model == "HOG-SVM":
            with open(f'./svm_{person.lower()}.pkl', 'rb') as f:
                svm = joblib.load(f)
            with open(f'./pca_{person.lower()}.pkl', 'rb') as f:
                pca = joblib.load(f)

            face = preprocess_image_svm(face)

            hog = extract_hog_features(face)

            hog_pca = pca.transform([hog])

            pred = svm.predict(hog_pca)

            if pred == 1:
                st.write("Match")
            else:
                st.write("Not Match")
    else:
        st.write("Face not detected")

def main():
    st.title("Real-time Face Verification App")
    
    model = st.selectbox("Select Model", ["Siamese", "HOG-SVM"])
    person = st.selectbox("Select Person", ["Theo"])
    enable = st.checkbox("Enable camera")
    captured_image = st.camera_input("Take a picture", disabled=not enable)

    if captured_image:
        verify(captured_image, model, person)

if __name__ == "__main__":
    main()