From Pixels to DNA: Why the Future of Compression Is About Every Kind of Data

2026-04-15 · Source: Towards Data Science · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Advanced, long

Summary

The landscape of data compression has expanded significantly beyond traditional audio and video, now encompassing diverse data types like genomes, point clouds, 3D scenes, and neural networks, driven by an exponential increase in global data generation, projected to reach 175 zettabytes by 2025. Organizations like ISO/IEC JTC 1/SC 29, responsible for JPEG and MPEG standards, are broadening their scope from "media for humans" to "data for humans and machines." Key developments include JPEG AI, a learning-based image coding standard using latent tensors for both human viewing and machine analysis; JPEG Trust, a framework for digital image authenticity and provenance; JPEG Pleno for plenoptic data like light fields and holograms; JPEG XS for ultra-low latency video; and JPEG DNA for storing images in biological molecules. MPEG continues to evolve video codecs, with the H.267 standard aiming for a ~40% bitrate reduction over VVC by 2028, while also prioritizing energy efficiency and integrating AI-based tools like Neural Network Video Coding (NNVC). Furthermore, new standards like Video Coding for Machines (VCM) and Feature Coding for Machines (FCM) are emerging to optimize visual data for machine tasks, with FCM offering up to 97% bandwidth savings by compressing intermediate neural features instead of pixels. Neural Network Coding (NNC) can compress deep neural networks by up to 97% without accuracy loss, and Gaussian Splat Coding (GSC) addresses 3D scene compression. Audio standards like MPEG-H Audio are moving towards object-based approaches for immersive and personalized experiences, including features like Dialog+ for enhanced speech clarity.

Key takeaway

For Computer Vision Engineers developing systems that process visual data, you should prioritize compression standards that support both human and machine consumption, such as JPEG AI and MPEG's VCM/FCM. These advancements enable significant bandwidth savings and enhanced privacy by processing intermediate features rather than raw pixels, while also integrating authenticity frameworks like JPEG Trust. Your adoption of these evolving standards will be crucial for building scalable, energy-efficient, and trustworthy AI-driven applications.

Key insights

Data compression is evolving into a foundational technology for all digital data, shifting from human-centric media to machine-centric semantics.

Principles

• Compression is foundational for digital ecosystems.
• Energy efficiency is a core codec selection criterion.
• Trust and authenticity are critical for digital media.

Method

JPEG AI transforms images into latent tensors for compression, enabling both human viewing and direct machine analysis. FCM compresses intermediate neural features from edge devices for cloud inference, reducing bandwidth and enhancing privacy.

In practice

• Use JPEG AI for images needing both human and machine interpretation.
• Implement JPEG Trust to verify digital image provenance.
• Consider FCM for M2M systems to reduce bandwidth by 97%.

Topics

• Universal Data Compression
• JPEG Standards
• Next-Gen Video Codecs
• AI-Enhanced Compression
• Machine Vision Data

Best for: Computer Vision Engineer, AI Engineer, Machine Learning Engineer, Research Scientist

Related on AIssential

• CVPR 2026 Papers with Code & Data — Publicly available code and data accelerate computer vision research and community engagement.
• What Matters in Practical Learned Image Compression — Optimizing learned image codecs for both perceptual quality and on-device speed yields superior compression and performance.
• Exploiting Semantic and Pixel Representations for Ultra-Low Bitrate Image Compression — Ultra-low bitrate image compression can achieve pixel-level fidelity by combining semantic and pixel representations in a diffusion model.
• Learned JPEG Compression for DNN Vision — J4D optimizes JPEG for DNNs using a differentiable codec and rate estimator, significantly boosting accuracy or reducing file size.
• Driving Safer AVs Faster with Smart Simulation, Neural Reconstruction, and Data-Centric Tools - Ep. 289 — Generative AI and neural reconstruction are accelerating AV simulation, enabling safer, more efficient autonomous driving development.
• Dual-Constrained Diffusion Image Compression for Operational Rate-Distortion-Perception Optimization — DCIC navigates the full RDP surface by integrating diffusion with dual distortion and idempotence constraints for flexible image compression.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards Data Science.