Peter works at Continental ADAS as Head of AI Data Group. He and his team are responsible for providing the right data in the right time for the AI development project teams which are working on perception function implementation. The task covers operative data management, tool / data pipeline implementation and prototyping of new methods. He joined Continental’s Artificial Intelligence Development Center at Budapest in 2018. Before that he worked for T-Systems where he participated in customer projects as a data engineer / architect for data warehousing, data integration and real-time data processing.
At Continental ADAS we are creating the next generation of software solutions to power safety-critical systems and enable autonomous driving features in new vehicles around the globe. These software components should continuously perform with high confidence in a wide range of scenarios, even in difficult light conditions, extreme weather, and complex environments. The necessary quantity, quality and distribution of ground truth samples are the fuel of the training process. Our task is to engineer the right datasets in the right time for the training, test, and validation of the CNN networks.
In this talk we will go through the different phases of an implementation project, how the requirements are changing over time as the algorithm performance improves. I will show you how we are preparing the dataset content in the first months of the project to have an initial version of the product. I will also give examples of the methods we are using in the fine tuning phase to improve the performance on domain specific scenarios. And finally you will see how we are building up an automatic data pipeline to continuously improve the performance on this long journey.
During the talk you will get an overview of the systems and tools we are working on to solve these data challenges.