Apple’s PICO cuts image bitrate 2.3-3x versus AV1 and VVC
Apple introduced PICO (Perceptual Image Codec), a learned image codec optimized for the human visual system that offers significant bitrate savings compared to existing codecs. PICO achieves 2.3-3x bitrate savings against AV1, AV2, VVC, ECM, and JPEG-AI, and 20-40% savings against other learned codecs, while maintaining fast on-device encoding and decoding speeds. The development was based on a comprehensive study of modeling choices and extensive subjective user studies to jointly optimize perceptual quality and runtime.
Key Takeaways
- PICO reports 2.3-3x bitrate savings against AV1, AV2, VVC, ECM and JPEG-AI.
- Apple says PICO beats other learned codec alternatives by 20-40% on bitrate.
- On an iPhone 17 Pro Max, PICO encodes 12MP images in 230ms and decodes them in 150ms.
- Apple says PICO was derived from millions of model configurations and large-scale subjective user studies.
- PICO comes with cross-platform robustness guarantees, unlike most learned codecs.
Why It Matters
Apple is showing that learned image compression can be both perceptually tuned and fast enough for on-device use, not just a research result. The benchmark claims matter because PICO is compared against AV1, AV2, VVC, ECM, JPEG-AI and other learned codecs, with subjective user studies backing the bitrate numbers. For the streaming stack, that makes image compression more relevant to mobile delivery and app-side asset handling. The next concrete signal to watch is whether Apple publishes the full arXiv paper and citation details behind PICO’s 2.3-3x savings and 230ms/150ms timing claims.
Read full article at apple.github.io