Apple’s PICO codec claims 2.3–3x bitrate savings over AV1
Apple has released PICO (Perceptual Image Codec), an AI-based still-image codec, along with an accompanying paper and code. PICO is designed for human visual quality and mobile-class runtime, aiming to make machine learning-based image compression practical for camera pipelines, photo storage, and cloud services. The codec claims significant bitrate savings over existing standards like AV1/AV2/VVC in subjective tests.
Key Takeaways
- Apple published PICO, a perceptual image codec, with an accompanying paper titled “What Matters in Practical Learned Image Compression.”
- PICO targets 12-MP images and iPhone-class runtime, encoding a 12-MP image in a few hundred milliseconds.
- Apple says PICO delivers 2.3–3x bitrate savings over AV1, AV2, VVC, and JPEG-AI in subjective tests.
- The release includes code, models, and detailed methodology, making the codec reproducible for camera pipelines, photo storage, and cloud services.
Why It Matters
PICO makes learned image compression look practical rather than purely experimental: Apple is pairing subjective quality gains with iPhone-class latency, which matters for camera pipelines, photo libraries, and cloud storage workflows. The competitive reference set is telling too, with Apple benchmarking against AV1, AV2, VVC, and JPEG-AI rather than just JPEG or HEVC-era formats. The key signal to watch is whether Apple’s published code and models lead to broader adoption or follow-on work around 12-MP, mobile-runtime learned codecs.
Read full article at linkedin.com