Image recognition programs require large sets of training data to produce accurate results. Human workers may categorize training sets that programs may use as training data to learn how to recognize objects. To increase the efficiency of the workers, it is proposed to break the categorization down into multiple steps in a pipeline. Different groups of workers will provide input at different stages of the pipeline, and the input from one group of workers will be passed to another group of workers. Breaking the categorization down into smaller tasks may increase the efficiency of the workers.
Garg, Rahul and Gowal, Sven, "Pipeline To Generate Training Data For Image Recognition", Technical Disclosure Commons, (March 10, 2015)