🐯虎嗅•Stalecollected in 14m
Apple's PICO: AI-driven image compression at 1/3 file size
💡Apple's new AI codec beats industry standards by 3x, proving perceptual optimization is the new frontier for media.
⚡ 30-Second TL;DR
What Changed
Achieves 30-43% of the file size of AV1/VVC/JPEG AI at equivalent visual quality.
Why It Matters
PICO demonstrates that prioritizing human perception over traditional PSNR metrics is the future of efficient media delivery, potentially setting a new standard for mobile image processing.
What To Do Next
Review the PICO paper on arXiv to understand how to implement perceptual loss functions in your own image generation or compression pipelines.
Who should care:Researchers & Academics
Key Points
- •Achieves 30-43% of the file size of AV1/VVC/JPEG AI at equivalent visual quality.
- •Uses a 'One-shot Context Model' to solve the speed bottleneck of autoregressive entropy coding.
- •Implements specialized loss functions (TextFidelityLoss, TilingArtifactLoss) to prevent GAN-induced hallucinations and tiling artifacts.
- •Real-time performance on iPhone 17 Pro Max: 230ms encoding and 150ms decoding for 12MP photos.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗