A Benchmark Dataset to Study the Representation of Food Images?