mage classification relies on labeled image data. Where does that come from? How accurate are the labels? How hard is it to create the labels? Read http://karpathy.github.io/2014/09/02/ what-i-learned-from-competing-against-a-convnet-on-imagenet/ to see how one machine learning researcher, Andrej Karpathy, spent time labeling ImageNet data and measured his own accuracy at 94%. Try the task yourself and see what accuracy you achieve. Can you think of better ways of collecting labels for images?