Data and Models Version control in Computer Vision meetup
The ML-REPA community invites you
to an open meeting on reproducibility and experiment management in computer vision, which will take place on August 15 at the Raiffeisenbank office in Nagatino.
At the mitap, we will deal with the features of ensuring reproducibility of experiments in Computer Vision, automation of pipelines and versioning of models. Where can DVC or MLFlow come in handy? And where is it better to write your “bike”? Also take a deeper look at the Catalyst implementation and its application.Full program and description of reports:
Vlad Grozin, Data Scientist at JOOMMLFlow for experiment reproducibility
In the report I will talk about MLFlow in the context of conducting experiments in CV. I will touch upon his philosophical differences between MLFlow and DVC, and about where it is worth using one, and where - the other.
Roman Tezikov, Middle Data Scientist at Dbrain, Catalyst Library DeveloperWokrshop: Dive into Catalyst
Consider a playable CV using the Catalyst framework. I will tell you how it works and what kind of goodies it has.
Dmitry Nikulin, Research Engineer, Samsung AI Center MoscowExperiment management in research: case study
A couple of days ago we published an article
at the intersection of CV and RL. I will tell you about the infrastructure written for this, how we set up the experiments and what bicycles we had to invent.
We open the doors for guests at 18:30, the start of the meeting at 19:00
Participation in the event is free, but by prior registration