Software engineering is a powerful but too often overlooked discipline that, when done well, can positively impact the development and implementation of computer vision algorithms. This tutorial will focus on the significant work required to design, develop, run and maintain a full software system around the types of algorithms we typically see presented at CVPR.  

We will shine a light on the underpublicized processes, practices, and common hurdles, as well as explain the complexity of large system development.  

Topics for this tutorial include training data creation, storage, and organization; neural networks and model storage; accuracy and regression testing in evolving environments; production deployment and multi-language development. 



Friday June 22, 2018

08:30 - 09:00 - Accelerating Algorithm Development, Evaluation, and Deployment by Providing Frameworks, David Doria (HERE)

09:00 - 09:30 - Reproducibility - An Industrial Perspective, Jan Ernst (Siemens)

09:30 - 10:00 - Scaling Active Learning for the Development of Imagery-Derived Maps, Ben Kadlec (Uber)

10:00 - 10:30 - Coffee break

10:30 - 11:00 - Building Computer Vision Systems with Open Source Software, Matt Turek (Kitware)

11:00 - 11:30 - Transforming Research Code Into Robust Multiplatform Mobile Products, Stephen Miller (Fyusion)

11:30 - 12:00 - Transfer Learning: Data Curation, Training, and Deployment Strategies, Tim Franklin (Microsoft)

12:00 - 12:30 - Research to Prod: Large Scale Visual Recognition in the Cloud., Wei Xia (Amazon)