Learning Large Scale Automatic Image Captioning