Alexey Shabanov: Metric Learning with Catalyst

Data fest Online 2020 Catalyst Workshop track Let’s imagine that you need to map your data points (images, texts, sounds and so on) to some vectors. For instance, in the case of working with images, the simplest solution would be to take the outputs of the penultimate layer of classifier. But what if you want to do something smarter? To solve such a problem, you can use set of tools from the field of Metric learning: sample triplets of data points using various strategies, use triplet loss, and calculate retrieval metrics. We have implemented all these tools for you in one place and provided a user-friendly interface so that you can easily adapt it to your task. In our video section, we will show with specific examples how the work with Metrics learning pipeline is arranged in Catalyst. Register and get access to the tracks: Join the community: