Event Preprocessing Tutorial

Stream of events is easy to be analyzed directly, but difficult to be used in modern deep learning models, which use tensors to accelerate the training. In order to benefit from the existing deep learning ecosystem, as well as to take full advantage of GPUs and other hardware accelerators, we enable encoding our raw events into tensors.

In these tutorials, you will learn about different preprocessing methods provided in the Metavision SDK API to extract key information out of the raw events, and to convert them to a dense representation that can be used in deep learning.

Preprocessing options:

• The C++ API provides facilities in the core module to preprocess events.

• Python bindings of these C++ classes allow for Python usage of these exact same procedures.

• The legacy Python API also provides full Python implementation of event preprocessors.

Currently, event preprocessing can be done in two ways in the Python API: the legacy one which is fully Python-based, and the new one which is based on bindings of the new C++ preprocessing classes. From now on, prefer this second version.