app.py

import gradio as gr
import numpy as np
import cv2
from tensorflow.keras.models import load_model

IMG_HEIGHT = 96
IMG_WIDTH = 96

# Load the saved Keras model
model = load_model("model_01.keras")

# Define the labels for ASL classes
labels = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I',
         'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S',
         'T', 'U', 'V', 'W', 'X', 'Y']  # Replace with your actual label names

def preprocess_frame(frame):
    """Preprocess the video frame for the ASL model."""
    # Convert the frame to a TensorFlow tensor
    if isinstance(frame, np.ndarray):
        frame = tf.convert_to_tensor(frame, dtype=tf.float32)
        # Reshape to add channel dimension if grayscale
        if frame.ndim == 2:  # If the input is grayscale
            frame = tf.expand_dims(frame, axis=-1)
            frame = tf.image.grayscale_to_rgb(frame)

    # Ensure the frame has 3 channels (RGB)
    if frame.shape[-1] == 1:  # Grayscale image
        frame = tf.image.grayscale_to_rgb(frame)

    # First scale down to dataset dimensions (if applicable)
    frame = tf.image.resize(frame, [28, 28])  # Resize to smaller dimensions for consistency

    # Resize to the target model input dimensions
    frame = tf.image.resize(frame, [IMG_HEIGHT, IMG_WIDTH])

    # Normalize pixel values to [0, 1]
    frame = tf.cast(frame, tf.float32) / 255.0

    # Add batch dimension for model input
    frame = tf.expand_dims(frame, axis=0)

    return frame


def predict_asl(frame):
    """Predict the ASL sign from the webcam frame."""
    # Preprocess the frame
    processed_frame = preprocess_frame(frame)
    # Make a prediction
    predictions = model.predict(processed_frame)
    # Get the class with the highest probability
    predicted_label = labels[np.argmax(predictions)]
    return predicted_label

css = """.my-group {max-width: 500px !important; max-height: 500px !important;}
            .my-column {display: flex !important; justify-content: center !important; align-items: center !important};"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_classes=["my-column"]):
        with gr.Group(elem_classes=["my-group"]):
            input_img = gr.Image(sources=["webcam"], type="numpy", streaming=True, label="Webcam Input")
            output_label = gr.Label(label="Predicted ASL Sign")

    input_img.stream(predict_asl, [input_img], [output_label], time_limit=30, stream_every=0.1)

demo.launch()