Machine Learning



Unlock the potential of Event-Based machine learning, with a set of dedicated tools providing everything you need to start execution of Deep Neural Network (DNN) with events.


The Metavision SDK ML module contains a collection of python utility modules to facilitate the navigation into large event-based datasets, and prepare everything you need to develop your first neural network:

  • metavision_ml.event_io : events recordings & live event-based camera manipulation.

  • metavision_ml.preprocessing: functions to apply to events before feeding them them to a machine learning model.

  • contains a class that allows to build a Sequential Dataloader for a training loop.

It also contains a set of C++ components that can be connected to form a live Detection & Tracking pipeline, working on a live events stream from a camera. The C++ API is using TorchJit to run inference with pytorch models.

A pre-built demo pipeline is delivered in the module, running inference on a pretrained automotive model.

Datasets & models

Access a large HD automotive dataset, with 15h of recordings and 23 millions of labels. Use our pytorch automotive model trained on this dataset.


You can already start looking in large event-based dataset, with our previously released QVGA dataset.


As part of the Metavision SDK ML module, run our Live detection and tracking pipeline, keeping track of targets in realtime at a 100Hz refresh rate.

More info

Want to know more, and start experimenting Machine Learning with Events?

Contact us and get access to Metavision Intelligence Professional.