Data cleansing
One of the most important part of the development of self-driving vehicles is large, reliable and properly labeled data. Data that meets these criteria can only be produced at the cost of hundreds or even thousands of hours of work by experienced people, and this time can grow exponentially as we want to create a database that meets more and more criteria. As part of the data cleansing project, we aim to minimize this cost and time-consuming process. As a solution, we have created a binary classifier neural network that can learn from the already validated database which of the newly processed data we consider valuable and which types of data are still missing. With this solution, we successfully reduced the time that required for data cleaning and partially automated the process of creating a learning database.
