Foundations of AI-Based Codec Development: An Introductory Course

Join Arsalan Zafar, Deep Render's CTO and co-founder, as he explains the fundamentals of AI-based video compression, demystifies traditional codec paradigms, and highlights the innovative approaches of Deep Render's codec technology.

Key Points:

• Arsalan introduces the concept of AI-based compression, distinguishing it from traditional codec methodologies.
• The discussion explores the evolution of AI and machine learning, focusing on their pivotal role in reshaping compression techniques.
• Arsalan explains the crucial components of AI-based compression, including architecture, loss function, and training processes.
• Arsalan details the intricate interplay between rate and distortion, essential for achieving optimal compression results.
• The video showcases practical applications of AI-based codecs in production environments, emphasizing their efficiency and scalability.

Contents:

00:00 Introduction - Arsalan introduces the concept of AI-based compression
00:01:22 Overview of AI and machine learning in compression
00:02:10 Exploring the architecture of AI-based codecs
00:03:41 Understanding the objective function and its role in compression
00:05:14 Defining rate and distortion in a differentiable manner
00:07:38 Differentiating distortion metrics and exploring implicit density losses
00:08:27 Illustrating the progression of compression throughout training
00:10:24 Transitioning from training to inference and practical implementation in production
00:12:48 Decoding AI-encoded video, and other features and advantages
00:15:28 Showcase of AI-based compression examples - from failures to successes
00:16:11 Conclusion and invitation to explore Deep Render's AI-based codec demonstrations.

For more on Deep Render, visit www.deeprender.ai.