load_process.py

import numpy as np
from keras import optimizers
import vggmodel
from keras.preprocessing import image as image_utils


class_names=["Stop navigation", "Excuse me", "I am sorry", "Thank you", "Good bye", "I love this grace", "Nice to meet you", "You are welcome", "How are you", "Have a good time", "Begin", "Choose", "Connection", "Navigation", "Next", "Previous", "Start", "Stop", "Hello", "Web"]
weights_path="model/weights-VggFinetune-17-0.65.f5"


def load_and_prcoess_image(path):
    image = image_utils.load_img(path, target_size=(175, 175))
    image = image_utils.img_to_array(image)
    input_image = np.expand_dims(image, axis=0)/255
    return input_image

model = vggmodel.create_model(175, 175)
# Load weights for model
model.load_weights(weights_path)

model.compile(loss='categorical_crossentropy',
              optimizer=optimizers.SGD(lr=0.00001, decay=1e-6, momentum=0.9),
              metrics=['accuracy'])

def predict_by_model(path):
    

    print("[INFO] loading and preprocessing image...")
    input_image = load_and_prcoess_image(path)
    prediction = model.predict(input_image)
    prediction_class = np.argmax(prediction, axis=1)

    if(is_confidence_too_low(prediction)):
        print("Can you say again? Please")
        write_to_txt("result_lip/text.txt", "Can you say again? Please")

    else:
        print(class_names[prediction_class[0]])
        print(prediction_class[0]+1)
        print(prediction[0])
        write_to_txt("result_lip/text.txt", class_names[prediction_class[0]])
    
    return prediction_class[0]