Sunday 10:50–11:25 in Megatorium

Half Hour of Flower: Planting the seeds for great Image Quality

Dirk Guijt

Audience level:
Intermediate

Description

Royal FloraHolland is the biggest flower auction in the world. And next to handling a lot of plants and flowers, we also handle a lot of pictures of plants and flowers. Pictures that need to be of a certain quality. We built a fully automized self-improving system to check the image quality by using deep learning models, asynchronous API technologies and Docker/AWS/Gitlab.

Abstract

Royal FloraHolland is the largest flower auction in the world! We handle 12 billion plants and flowers every year. We also handle the millions of pictures of plants and flowers that are used in the auctioning process. And these pictures need to meet our photo requirements and quality standards. In this talk I will show our really cool (it's awesome) infrastructure that we, the Data Science & Engineering team, built to automatically check the photo quality of every photo. We have used some deep learning models, some heuristics, and an asynchronous micro-service API infrastructure to process all the requests for our various use cases. I will go through the whole setup and explain how we got from just a single deep neural net to a fully productionized self-improving system.

Subscribe to Receive PyData Updates

Subscribe