import cv2
import numpy as np
import gradio as gr
from mtcnn import MTCNN
from tensorflow.keras.models import load_model
from tensorflow.keras.applications.xception import preprocess_input as xcp_pre
from tensorflow.keras.applications.efficientnet import preprocess_input as eff_pre
from huggingface_hub import hf_hub_download

# Download and load models
xcp_path = hf_hub_download(repo_id="Zeyadd-Mostaffa/deepfake-image-detector_final", filename="xception_model.h5")
eff_path = hf_hub_download(repo_id="Zeyadd-Mostaffa/deepfake-image-detector_final", filename="efficientnet_model.h5")
xcp_model = load_model(xcp_path)
eff_model = load_model(eff_path)

# Load face detector
detector = MTCNN()

# Detection thresholds
MIN_FACE_SIZE = 60        # Accept faces larger than 60×60 pixels
MIN_CONFIDENCE = 0.94     # Accept only confident detections

def predict(image):
    faces = detector.detect_faces(image)
    if not faces:
        return "No face detected", image

    output_image = image.copy()
    results = []
    valid_faces = 0

    for idx, face in enumerate(faces):
        conf = face.get("confidence", 0)
        x, y, w, h = face['box']

        # Filter out unclear faces
        if w < MIN_FACE_SIZE or h < MIN_FACE_SIZE or conf < MIN_CONFIDENCE:
            continue

        img_h, img_w = image.shape[:2]
        margin = 0.2
        x = max(0, int(x - w * margin))
        y = max(0, int(y - h * margin))
        x2 = min(img_w, x + int(w * (1 + 2 * margin)))
        y2 = min(img_h, y + int(h * (1 + 2 * margin)))

        face_img = image[y:y2, x:x2]

        # Resize and preprocess
        face_xcp = cv2.resize(face_img, (299, 299))
        face_eff = cv2.resize(face_img, (224, 224))
        xcp_tensor = xcp_pre(face_xcp.astype(np.float32))[np.newaxis, ...]
        eff_tensor = eff_pre(face_eff.astype(np.float32))[np.newaxis, ...]

        # Predict
        pred_xcp = xcp_model.predict(xcp_tensor, verbose=0).flatten()[0]
        pred_eff = eff_model.predict(eff_tensor, verbose=0).flatten()[0]
        avg = (pred_xcp + pred_eff) / 2

        label = "Real" if avg > 0.41 else "Fake"
        color = (0, 255, 0) if label == "Real" else (0, 0, 255)

        # Draw on image
        cv2.rectangle(output_image, (x, y), (x2, y2), color, 2)
        cv2.putText(output_image, f"{label} ({avg:.2f})", (x, y - 10),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.6, color, 2)

        results.append(f"Face {idx+1}: {label} (Avg: {avg:.3f}, XCP: {pred_xcp:.3f}, EFF: {pred_eff:.3f})")
        valid_faces += 1

    if valid_faces == 0:
        return "No clear or confident face detected", image

    return "\n".join(results), output_image

# Interface
interface = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="numpy", label="Upload Image"),
    outputs=[
        gr.Textbox(label="Predictions"),
        gr.Image(type="numpy", label="Annotated Image"),
    ],
    title="Deepfake Detector (Multi-Face Ensemble)",
    description="Detects all visible faces in an image and classifies each as Real or Fake using Xception and EfficientNetB4 ensemble.",
)

interface.launch()