Report on the JPEG AI Call for Evidence Results
In the context of the JPEG AI learning-based image coding activity, a Call for Evidence (CfE) was issued as an outcome of the 86th JPEG meeting, Sydney, Australia for an image coding system that achieves substantially better compression efficiency than existing solutions, namely by exploiting advanced machine learning tools, such as deep neural networks. Four submissions were received in response to the CfE.
During the 89th JPEG meeting, the CfE submissions on learning-based image coding were presented and discussed along with the results of their subjective evaluation. This document reports both the objective and subjective performance evaluation as well as the coding complexity of all CfE proponents submissions, this means the four teams which have made submissions for the call. Additionally, the performance evaluation of two teams which have submitted for the learning-based image coding challenge organized at IEEE 22nd International Workshop on Multimedia are also included.
To stay informed on JPEG AI activities, please regularly consult our website at jpeg.org and/or subscribe to our e-mail reflector.