Skip to content

Introduction to Visual Backprop

As part of monitoring and managing your experiments, you might wish to visualize your data. Visualizing data assists in analyzing the experiment when solving non-classification problems such as the angle of a steering wheel, based on an image provided to the model.

Visual Backprop is an algorithm designed by Nvidia for using heat maps to visualize which part of the image provided to a model is important to it.

By using Visual Backprop with MissingLink, the dashboard can assist you in tuning the model by showing the heat maps. This functionality can be combined with the test monitoring in allowing a drill down to the actual data points with their respective visualizations.

If you have performed basic integration with the MissingLink SDK, learn now how to generate Visual Backprop for the supported frameworks: