Data-Centric Computer Vision

Approaches, Applications, and Best Practices using the Appen Data Annotation Platform

Human vision systems have the advantage of learning from a lifetime of experiences how to contextualize the things we see. Machine learning models, on the other hand, usually need a substantial number of real-world scenarios to learn from to be able to product reliable computer vision (CV) outputs. These examples may come in many forms:

• 2D images and video (taken from an SLR or infrared camera)
• 3-D images and video (taken from a camera or scanner)
• Sensor data (taken from RADAR or LiDAR technology).
• Sometimes a mix of the above

High-quality data is foundational to effective CV systems. Today, with the right training data, a Computer Vision system can recognize objects in images and video, including their shapes, textures, colors, sizes, locations, movements, and other relevant characteristics.

This Computer Vision eBook is focused on a data-centric approach to model development, which consists of systematically changing and enhancing datasets to improve output accuracy, as opposed to adjusting the models.

The Data-Centric Computer Vision eBook Will Cover:

• Approaches and applications for different computer vision use cases
• Techniques and best practices for launching successful CV-based models
• Visual examples using our platform and output code