In the early days of autonomous vehicle development, training data projects primarily employ basic data annotation tools, such as bounding boxes and polygon identification. Over time, data segmentation requirements grow more complex and detailed.
As the infographic illustrates, the increasing complexity of each task results in exponentially more time and effort – and skills – to deliver the level of training data the algorithm requires to think and see like a human.
The ascending level of complexity explains why AV developers have come to rely on third-party data annotation firms. As with other specialized tasks that require an advanced, highly trained workforce and specialized tools, data annotation has evolved into its own Machine Learning niche.