import time
import os
import cv2
import numpy as np
import mediapipe as mp
import tensorflow as tf
from flask import Flask, request, jsonify
from flask_cors import CORS

app = Flask(__name__)
CORS(app)

# Load model dari folder yang sama
MODEL_PATH = "best_sibi_landmark.keras" 
print("[INFO] Memuat model Keras...")
model = tf.keras.models.load_model(MODEL_PATH)
classes = list("ABCDEFGHIJKLMNOPQRSTUVWXYZ")

mp_hands = mp.solutions.hands
hands = mp_hands.Hands(static_image_mode=True, max_num_hands=2, min_detection_confidence=0.5)

def normalize(points):
    points = np.array(points)
    wrist = points[0]
    points[:, 0] -= wrist[0]
    points[:, 1] -= wrist[1]
    points[:, 2] -= wrist[2]
    return points.flatten()

@app.route('/predict', methods=['POST'])
def predict():
    # Catat waktu mulai dari awal request masuk
    start_time = time.time()
    try:
        file = request.files['image']
        npimg = np.frombuffer(file.read(), np.uint8)
        img = cv2.imdecode(npimg, cv2.IMREAD_COLOR)

        rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        result = hands.process(rgb)

        # Hitung durasi sementara jika tangan tidak terdeteksi
        end_time = time.time()
        processing_time = round((end_time - start_time) * 1000)

        if not result.multi_hand_landmarks:
            return jsonify({
                "status": "error", 
                "message": "Tangan tidak terdeteksi!",
                "latency": f"{processing_time} ms"
            })

        left = [0] * 63
        right = [0] * 63

        for idx, hand in enumerate(result.multi_hand_landmarks):
            pts = [[lm.x, lm.y, lm.z] for lm in hand.landmark]
            pts = normalize(pts)
            hand_type = result.multi_handedness[idx].classification[0].label
            if hand_type == "Left":
                left = pts.tolist()
            else:
                right = pts.tolist()

        feature = np.array(right + left, dtype=np.float32).reshape(1, 126, 1)
        pred = model.predict(feature, verbose=0)[0]
        best_idx = np.argmax(pred)
        
        # Hitung total waktu proses sampai prediksi selesai
        end_time = time.time()
        processing_time = round((end_time - start_time) * 1000)

        return jsonify({
            "status": "success",
            "letter": classes[best_idx],
            "confidence": f"{pred[best_idx] * 100:.2f}%",
            "latency": f"{processing_time} ms"
        })
        
    except Exception as e:
        end_time = time.time()
        processing_time = round((end_time - start_time) * 1000)
        return jsonify({
            "status": "error", 
            "message": str(e),
            "latency": f"{processing_time} ms"
        })

if __name__ == '__main__':
    app.run(port=5000, debug=True)