Video/Audio Quality in Computer Vision Workshop @ WACV2023

Gerard Medioni

Vice president and distinguished scientist of Physical Stores Tech organization, Amazon.com

https://www.amazon.science/author/gerard-medioni

Title: Inventing at Amazon for Physical Stores

Abstract:

Through 3 examples from the Physical Stores domain, I will discuss the process of invention at Amazon.

We will take a look under the hood and discuss the technology powering

Just Walk Out: the retail experience which lets shoppers enter a store, grab what they want, and get going.

Amazon One: from image acquisition to matching, the fast, convenient, contactless identity service that uses your palm – just hover to enter, identify, and pay; simply by being you.

Amazon Style: Amazon's first-ever physical store with clothing, shoes, and accessories for men, women, and kids.

Sriram Sethuraman

Sr. Principal Scientist in the Prime Video playback organization, Amazon.com

https://www.amazon.science/author/sriram-sethuraman

Title: Next level challenges in Video Quality Understanding

Abstract: Understanding the perceived quality of a video is required for progressively optimizing compressed visual representations while maintaining a high user experience. Multiple ML-based ratings-driven video quality models have been developed. However, the holy grail continues to be a model that accurately predicts the threshold behavior in human vision where differences start to be perceived beyond a certain content-specific distortion level. This talk will summarize some of the progress in this space in the recent past, including some of our team’s efforts with academic partners, through the lens of the validity of these models for encoding optimization. Some of the next level challenges that such models struggle with, and some initial thoughts on how task-specific models can augment the understanding will also be presented.