AI BriefingPerceptual Image Codec: The Key to Practical Learned Image Compression
·2026.05.24 21:01
Key point
Introducing PICO, which is optimized for the human visual system and dramatically improves compression ratio and speed compared to existing codecs.
Details
PICO (Perceptual Image Codec) is the first practical learned codec optimized directly for the Human Visual System (HVS). By exploring millions of model configurations, it simultaneously optimizes Perceptual quality and On-device runtime.
The key performance and features are as follows:
- Compression efficiency: It achieves a 2.3-3x bitrate savings compared to existing codecs such as AV1, VVC, and JPEG-AI, and shows 20-40% higher efficiency compared to the best existing learned codecs.
- Runtime speed: On an iPhone 17 Pro Max, it can encode a 12MP image within 230ms and decode within 150ms. This is faster than most ML-based codecs running on a V100 GPU.
- Robustness: It has Cross-platform robustness, ensuring consistent performance across various platforms.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Cactus Compute Releases Ultra-Compact 'Needle 3' Model (29MB), Claims DeepSeek V4 Flash-Level PerformancePPinterest Unveils 3-Tower Engagement Co-Train Model: Achieves 1% CTR Lift and 30ms Latency