predict_image.py

# preprocess data
(data, sizes) = preprocess(img_rdd, means, scale, resolution)

# image prediction
output = model.predict(data)

# scale normalized results according to original image sizes
result = output.zip(sizes).map(lambda (output_i, img_size):
    scaleOutput(output_i, img_size)
 ).collect()

# visualize detections
for img_id in range(len(result)):
    detections = result[img_id]
    visualize(images[img_id], detections)
    plt.subplots(figsize=(6, 6))
    plt.imshow(cv2.cvtColor(images[img_id], cv2.COLOR_BGR2RGB))