ff94e24d202ba1e5861caddc915df0c43f23b766,examples/application_object_detection.py,,main,#,62

Before Change

    gradients = frcnn.loss_gradient(x=image, y=None)

    // Create adversarial image
    image_adv = image + np.sign(gradients) * 8 * 
    image_adv = np.clip(image_adv, a_min=0, a_max=255).astype(np.uint8)

    for i in range(image_adv.shape[0]):
        plt.axis("off")
        plt.title("image_adv {}".format(i))

After Change

def main():
    // Create object detector
    frcnn = PyTorchFasterRCNN(clip_values=(0, 255))

    // Load image 1
    image_1 = cv2.imread("./10best-cars-group-cropped-1542126037.jpg")
    image_1 = cv2.cvtColor(image_1, cv2.COLOR_BGR2RGB)  // Convert to RGB
    print("image_1.shape:", image_1.shape)

    // Load image 2
    // image_2 = cv2.imread("./banner-diverse-group-of-people-2.jpg")
    // image_2 = cv2.cvtColor(image_2, cv2.COLOR_BGR2RGB)  // Convert to RGB
    // print("image_2.shape:", image_2.shape)

    // Stack images
    image = np.stack([image_1], axis=0)
    print("image.shape:", image.shape)

    for i in range(image.shape[0]):
        plt.axis("off")
        plt.title("image {}".format(i))
        plt.imshow(image[i], interpolation="nearest")
        plt.show()

    // Make prediction on benign samples
    predictions = frcnn.predict(x=image)

    // Process predictions
    predictions_class, predictions_boxes, predictions_class = extract_predictions(predictions[0])

    // Plot predictions
    plot_image_with_boxes(img=image[0].copy(), boxes=predictions_boxes, pred_cls=predictions_class)

    // // Calculate loss gradients
    // gradients = frcnn.loss_gradient(x=image, y=None)
    //
    // // Create adversarial image
    // image_adv = image + np.sign(gradients) * 8 * 1
    // image_adv = np.clip(image_adv, a_min=0, a_max=255).astype(np.uint8)

    attack = FastGradientMethod(classifier=frcnn, eps=8)
    image_adv = attack.generate(x=image, y=None)

    print("np.amax(np.abs(image - image_adv))")
    print(np.amax(np.abs(image - image_adv)))

Instances