Presented By O’Reilly and Cloudera
Make Data Work
21–22 May 2018: Training
22–24 May 2018: Tutorials & Conference
London, UK

Deep computer vision for manufacturing

Aurélien Géron (Kiwisoft)
14:5515:35 Wednesday, 23 May 2018
Average rating: ***..
(3.67, 3 ratings)

Who is this presentation for?

  • Anyone interested in computer vision

Prerequisite knowledge

  • A basic understanding of machine learning concepts (e.g., what machine learning is, what a training set is, and what it means to train a model)

What you'll learn

  • Understand the state-of-the-art deep computer vision architectures (CNNs) and their applications, particularly in manufacturing

Description

Computer vision in manufacturing has actually been around for decades: it’s present in thousands of production lines, performing product classification, detecting defective items, gathering data for analytics, and more. Very recently, companies have started to shift from classical computer vision techniques to modern techniques based on deep learning, namely convolutional neural networks (CNNs), which can achieve amazing precision, often reaching or even exceeding human abilities.

Aurélien Géron details common CNN architectures for classification (e.g., ResNet), image segmentation (e.g., DeepLab), object detection (e.g., YOLO), and anomaly detection (e.g., ResNet+SVM), explains how they can be applied to manufacturing, and covers potential challenges along the way, including:

  • Having to gather and possibly label a training set for each new product type;
  • Difficulties in interpreting the system’s decisions;
  • Avoiding model rot;
  • Handling high volume and low latency.
Photo of Aurélien Géron

Aurélien Géron

Kiwisoft

Aurélien Géron is a machine learning consultant at Kiwisoft. Previously, he led YouTube’s video classification team and was founder and CTO of two successful companies (a telco operator and a strategy firm). Aurélien is the author of several technical books, including the O’Reilly book Hands-on Machine Learning with Scikit-Learn and TensorFlow.