CALL FOR PARTICIPATION

JPEG AI Learning-based Image Coding Challenge (http://jpegai.github.io)

At 22nd IEEE International Workshop on Multimedia Signal Processing (MMSP) 2020 (https://attend.ieee.org/mmsp-2020/)

JPEG AI Learning-based Image Coding Challenge solicits technical contributions that demonstrate efficient coding of image content based on machine learning approach. This means image coding solutions for which learning-based modules (end-to-end trained) play an important role, considering the overall codec architecture. Some examples are non-linear data correlation transformations, learned probability models for entropy coding, hierarchical learning-based coding structures and perceptual optimizations. 

The Challenge is designed to be in sync with a JPEG AI Call for Evidence (CfE) on Learning-based Image Coding (https://jpeg.org/items/20200224_cfe_peg_ai.html) and will use the same content, evaluation methodologies and deadlines. The objective of the above CfE is to objectively and subjectively evaluate relevant learning-based image coding solutions to validate the potential performance of this novel coding approach, especially in terms of compression efficiency.

Participants are required to submit the following materials:

* A detailed description of the coding algorithm in the form of an MMSP challenge paper and/or technical report to JPEG as well as methodologies and data used for training.

* A decoder implementation in a form that allows stand-alone inference/testing on a standard computer (CPU only) in a reasonable amount of time, preferably in source code form.

* Compressed codestreams and corresponding decoded images.

Timeline

30th May: Proposal registration.

12th June: Submission of code-streams, decoded images, MMSP challenge paper and/or JPEG technical document.

14th June: Release of the test dataset for proponents to code.

18th June: Submission of code-streams and decoded images for the test dataset.

21st June: MMSP Challenge paper and/or JPEG technical document submission.

15th July: MMSP Challenge paper notification.

30th July: MMSP Challenge camera ready paper.

1st September: 1st Stage: Objective evaluation of all the proposals; 2nd Stage: Subjective evaluation.

15th September: End of subjective evaluation; all results will be released online.

21-23rd September: Challenge session at MMSP 2020 with presentations and announcement of award recipients.

89th JPEG Meeting: Presentation and discussion of the proposals at the JPEG meeting (planned for 10-16 October)

The MMSP Challenge paper submissions will be peer reviewed in a similar way as any other regular MMSP paper submissions and, if accepted, will be included in the IEEExplore.

Based on the experimental results obtained and the review of the significance, novelty and impact of the submitted image coding solutions, two awards will be discerned:

* Best performing image coding solution.

* Most innovative image coding solution.

E-mail reflector for discussions and questions: jpeg-ai@jpeglists.org 

To subscribe to the reflector please visit http://jpeg-ai-list.jpeg.org or in case of problems contact lists@jpeg.org.

Kind regards,

João Ascenso (joao.ascenso@lx.it.pt)

Fernando Pereira (fp@lx.it.pt)

Touradj Ebrahimi (touradj.ebrahimi@epfl.ch)