Exploration on JPEG AI
JPEG Committee created the learning-based image coding (JPEG AI) ad hoc group which aims to assess and develop image coding solutions following an end-to-end learning based coding approach, potentially towards the definition of a standard. The work carried out thus far covers the definition of training and testing datasets, benchmarking state-of-the-art codecs, coding conditions and a set of reliable objective quality metrics and subjective assessment procedures. The effort exhaustively evaluated multiple aspects of learning-based image coding to fully understand its strengths and weaknesses, notably when compared to traditional image coding technology.
Recently, JPEG AI has issued a Call for Evidence in connection with a challenge on AI based image compression. The goal of this activity is to objectively and subjectively evaluate relevant learning-based image coding solutions to demonstrate the potential of learning-based coding technologies. Moreover, JPEG AI is looking for evidence that learning-based image compression technologies can perform better than classic codecs, e.g. JPEG, JPEG2000, HEVC.