Perceptual Image Codec: What Matters in Practical Learned Image Compression (apple.github.io)

Researchers have introduced PICO (Perceptual Image Codec), a groundbreaking learned image compression codec that is both practical and optimized specifically for the human visual system. After a comprehensive analysis of various modeling choices and an extensive search over millions of configurations, PICO achieves a remarkable 2.3-3× bitrate reduction compared to traditional codecs like AV1, AV2, VVC, ECM, and JPEG-AI, alongside 20-40% savings against other leading learned codecs. The codec's efficiency is underscored by its ability to encode 12MP images on an iPhone 17 Pro Max in just 230ms and decode them in 150ms, significantly outpacing many existing ML-based codecs that typically require high-performance GPUs. This development is significant for the AI/ML community as it not only enhances image compression efficiency but also promises cross-platform robustness, making it versatile for various applications. PICO’s focus on perceptual quality ensures that users experience minimal degradation in visual fidelity, which is a critical factor for industries reliant on image quality such as media, gaming, and telecommunications. The combination of performance and perceptual optimization positions PICO as a strong contender in the evolving landscape of image compression technologies.