Hi!

[pentadotddot]

Hi!

Is it possible to just run inference on a random image, like python app.py --image ./test.jpg --model custom_cnn ? Also could you please provide the .h5 files? Nice and clean implementation though, thanks for tis repo!

[Sanjana7395]

This is the code to run inference on a random image:

def plot(image, model, label, x_hog=None):
    plt.figure(figsize=(8, 6))
    plt.imshow(image)
    plt.axis('off')

    predictions, percent = model_predict(model, image, x_hog, label)
    print(predictions)
    print(percent)

    plt.suptitle('Predicted label: {} ({:.2f} %)'.format(predictions[0], np.max(percent[0]) * 100))

    make_folder('results/visualization')
    plt.savefig('results/visualization/app.png', bbox_inches='tight')

def model_predict(model, x, hog, y):
    predictions = []
    percent = []
    labeler = LabelEncoder()
    labeler.fit(y)
    if model == "custom_cnn":
        custom_model = load_model(os.path.join(ROOT_DIR, 'custom.h5'))

        x = np.expand_dims(x, axis=0)
        percent = custom_model.predict(x)
        predictions = np.argmax(percent, axis=-1)
        predictions = labeler.inverse_transform(predictions)
        
    return predictions, percent


def main():
    """ Predict the disease of the given image.

    Usage example:
        python app.py -m custom_cnn -i path/to/image

    """
    # construct the argument parser and parse the arguments
    ap = argparse.ArgumentParser()
    ap.add_argument("-i", "--image_index", type=int, help="index of the test image")
    ap.add_argument("-img", "--image", type=str, help="index of the test image")
    ap.add_argument("-m", "--model", type=str, required=True, choices=("vgg", "custom_cnn", "svm", random_forest", "majority_voting", "stacked_prediction"), help="model to be used")
    args = vars(ap.parse_args())

    X_image = np.load('data/test/ImageTest_input.npy')
    y = np.load('data/test/DiseaseTest_input.npy')

    if args["image_index"]:
        image = X_image[args["image_index"]]

    elif args["image"]:
        path = args["image"]
        image = plt.imread(path)
        image = cv2.resize(image, (180, 180), cv2.INTER_AREA)

    plot(image, args["model"], y)


if __name__ == "__main__":
    main()

Also I have attached the CNN-Custom trained model
custom.zip

Hope this helps!

[pentadotddot]

Thank you it really helped me! I get you back with some results:) Thanks again!