Pik is a new lossy image format for the internet. This directory contains an encoder and a decoder for the format.
This project is in the initial research stage, please don't use it for any purpose.
Pik is to have roughly the same performance requirements space that JPEG is holding today. We aim to have roughly similar decoding speed, i.e., no more than 50 % slower in decoding, but adding more modern compression and prediction methods, giving 55 % more density. We plan to improve to 65 % before freezing the format. Some of these new advances may come with a moderate drop in decoding speed (possibly 40 % of jpeg speed).
The bitstream supports tiles so that multi-core decoding can decode a single image faster (up to 1 GB/s).
We are planning to keep the format 8x8 DCT based, possibly with some support for non-integral-transform-based direct mode blocks (or overlay blocks).
The software currently requires an AVX2 and FMA capable CPU, e.g. Haswell.
Please run git submodule init && git submodule update, then make -j8, which
creates cpik and dpik binaries in bin/.
The --distance command line argument to cpik is a Butteraugli distance (see
http://github.com/google/butteraugli), which indicates the largest acceptable
error. Larger values lead to smaller files and lower quality. Try 1.0 for a
visually lossless result.
- Butteraugli (HVS-aware image differences)
- Guetzli (JPEG encoder with denser packing)
This is not an official Google product.