Autonomous vehicle fleets and models generate hundreds of petabytes of real-world and synthetic video data, stressing edge throughput, GPU-based training and validation clusters, and long-term storage. This session will explore ML-aware, content-adaptive video data compression spanning from ingestion to GPU-accelerated cloud encoding, covering all pipeline needs, along with methodologies of sensitivity analysis to quantify downstream model impact. By optimizing ingestion, storage and pipeline throughput for real-world and synthetic footage, teams can reduce costs while preserving model fidelity and safety-critical performance.

• AV data challenges: managing hundreds of petabytes impacting infrastructure investments, network bottlenecks and costs
• Discover how ML-aware video data compression accelerates autonomous vehicle pipelines while preserving downstream model performance
• Explore validation framework: step-by-step protocol for testing compression impact on various stages in autonomous vehicle models
• AV video has various target end points with competing requirements. Learn about how to best tailor video processing to each