The researchers released a massive open-source dataset for others to use.

A clean, well-labelled dataset is essential for training accurate models.