import gradio as gr
import cv2
import mediapipe as mp
import numpy as np
import pandas as pd
import tensorflow as tf
import joblib
from sklearn.preprocessing import LabelEncoder

# ------------------ Load Model ------------------
model = tf.keras.models.load_model("model.h5")
scaler = joblib.load("scaler.pkl")

# ------------------ Load Labels ------------------
df = pd.read_csv(
    "keypoints_dataset+space.csv",
    low_memory=False
)

encoder = LabelEncoder()
encoder.fit(df["label"].astype(str))

# ------------------ Arabic Mapping ------------------
label_to_arabic = {
    "aleff": "ا",
    "bb": "ب",
    "ta": "ت",
    "thaa": "ث",
    "jeem": "ج",
    "haa": "ح",
    "khaa": "خ",
    "dal": "د",
    "thal": "ذ",
    "ra": "ر",
    "zay": "ز",
    "seen": "س",
    "sheen": "ش",
    "saad": "ص",
    "dhad": "ض",
    "taa": "ط",
    "dha": "ظ",
    "ain": "ع",
    "ghain": "غ",
    "fa": "ف",
    "gaaf": "ق",
    "kaaf": "ك",
    "laam": "ل",
    "meem": "م",
    "nun": "ن",
    "ha": "ه",
    "waw": "و",
    "yaa": "ي",
    "0": "0",
    "1": "1",
    "2": "2",
    "3": "3",
    "4": "4",
    "5": "5",
    "6": "6",
    "7": "7",
    "8": "8",
    "9": "9",
    "10": "10",
    "space": " "
}

# ------------------ MediaPipe ------------------
mp_hands = mp.solutions.hands

hands = mp_hands.Hands(
    static_image_mode=True,
    max_num_hands=1,
    min_detection_confidence=0.5
)

# ------------------ Prediction Function ------------------
def predict(image):

    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)

    results = hands.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))

    if not results.multi_hand_landmarks:
        return "No hand detected"

    hand_landmarks = results.multi_hand_landmarks[0]

    landmarks = np.array([
        [lm.x, lm.y, lm.z]
        for lm in hand_landmarks.landmark
    ])

    landmarks = landmarks.flatten().reshape(1, -1)

    landmarks = scaler.transform(landmarks)

    prediction = model.predict(landmarks, verbose=0)

    predicted_class = np.argmax(prediction)

    predicted_label = encoder.inverse_transform(
        [predicted_class]
    )[0]

    arabic_output = label_to_arabic.get(
        predicted_label,
        predicted_label
    )

    return f"Prediction: {arabic_output}"

# ------------------ Gradio App ------------------
app = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="numpy"),
    outputs="text",
    title="Arabic Sign Language Recognition"
)

app.launch()