Title: Improving Netflix Video Quality with Neural Networks

Introduction

This article discusses how neural networks can be used to improve Netflix video quality. Conventional video downscaling is replaced with a neural network-based approach, known as the deep downscaler.

NN-based Video Downscaling

The deep downscaler is a neural network architecture that improves video quality by learning a higher-quality video downscaler. The training approach is intuitive and produces a downscaler that is not tied to a specific encoder or encoding implementation.

Improved Netflix Video Quality

In preference-based visual tests, the deep downscaler was preferred by ~77% of test subjects across a wide range of encoding recipes and upscaling algorithms. A/B testing showed the benefit of deploying the deep downscaler for all devices streaming Netflix without playback risks or quality degradation.

Applying Neural Networks at Scale

To reduce input channels, we apply NN-based scaling on luma and scale chroma with a standard Lanczos filter. The deep downscaler is integrated into our next-generation encoding platform, Cosmos, and applied prior to encoding.

Future Work

We are exploring other use cases for neural networks, such as video denoising. Additionally, we are investigating more efficient solutions to applying neural networks at scale.