Training

Pre-trained networks are good for starting developing your applications and testing. However, nothing gives a boost to your performance as performing your own training on your own application-specific data. In this section, we will see how to train a network using event-based data.

There are two main ways of training your network:

• supervised training requires labeled data, also known as Ground Truth (GT)

• un-supervised training does not need labels and can learn directly from patterns in the data.

Depending on the type of training you want to use, follow one of our tutorials:

• object-detection network trained with supervised learning
• optical-flow network trained with un-supervised learning flow_trainer

Note

Obtaining the best results when performing your own training requires a deep knowledge of both machine learning and your application domain. Often, the best results are obtained by a trial-and-error process: experiment with different combinations of parameters, collect more data, choose a different pre-processing or noise filtering, etc.

Explore some more tutorials about training:

• Using a Sequential DataLoader to Create a Training Loop
• Pytorch RNN API

Note

Tutorials in this section were created using Jupiter Notebooks. You can execute them on your computer by downloading the source code at the top or bottom of the page. More information can be found on this page.