PEVC: Practical end-to-end video compression challenge in MMSP 2026

        1.  Description

Over the past five years, end-to-end video compression has been a research focus of image processing in both academia and industry. Several widely adopted learned image compression techniques, such as auto-encoders and probability estimation neural networks, have been successfully extended to learned video compression. Concurrently, the methodological pipeline of learned video compression has evolved from a residual-based framework to a condition-based one. Until recently, the performances of end-to-end video compression schemes have been claimed to surpass that of the H.266/(VVC) or even ECM. However, most methods do not use the same quality metric as H.266/(VVC) or ECM, which makes the reported results less convincing. Therefore, this challenge adopts the YUV-PSNR between the original and reconstructed videos in YUV420 formats as the quality metric and calls for proposals that can outperform ECM under a fair comparison.

Building upon the previous challenge held last year, this year’s edition further introduces key changes to steer the field toward practical applications. Specifically, to address the requirements of live streaming scenarios where encoding and decoding must be balanced, this challenge constrains both encoder and decoder complexities to 200 KMACs/pixel, encouraging hardware-friendly solutions.

In this challenge, the participants are required to compress all videos defined in the Test Dataset. The actual bitrate is not allowed to exceed a target bitrate (kbps), which is set to the bitrate of the test sequence coded by ECM using the quantization parameter 27 under the default random-access configuration. The organizers will provide the anchor results during the model test stage.

In the following sections, detailed information about the rules of participation, evaluation, dataset, and deadlines will be provided.

2. The Grand Challenge's Importance and Relevance to MMSP Workshop

The grand challenge on end-to-end video compression is crucial as it establishes a fair and practical benchmark for the field. By mandating a standardized quality metric and constraining computational complexity, it drives research towards hardware-friendly solutions that can surpass ECM. This focus directly aligns with the core themes of the MMSP workshop.

3. Rules for participation

Participants are invited to submit solutions for end-to-end video compression, following these rules:

l  The participants are requested to submit a decoder along with a docker environment and the corresponding script which can run the decoder.

l  The participants are requested to submit the compressed bitstreams. The bitstreams shall be named like I01.bin.

l  The participants are requested to submit the decoded videos. The decoded videos shall be named like I01dec.yuv.

l  The participants are requested to submit their grand challenge paper describing the solution using the MMSP format.

l  The top five participants are requested to give a presentation of their solutions at the Special Session organized at MMSP 2026.

 

        4. Evaluation

The decoded video sequences will be evaluated in YUV 4:2:0 color space. The weighted average PSNR=(6×PSNR_Y+PSNR_U+PSNR_V)/8  of the Y, U, and V components will be used to evaluate the distortion of the decoded video sequences. An anchor of ECM-14.0 coded with QP=27 under random access configuration defined in the ECM common test conditions (encoder_randomaccess_ecm.cfg) will be provided. The actual bitrate (kbps) of the bitstream of each video sequence is not allowed to exceed the target kbps of the test video coded by the anchor. The intra period in the proposed submission shall be no larger than that used by the anchor. The performance Q is evaluated by the weighted PSNR of the Y, U, and V components.

 

        5. Dataset

l  Training and Validation Dataset: It is recommended to use the UVG and CDVL dataset for training. Participates are free to split the provided videos into training and validation dataset. Participates are also free to use some other dataset for training and validation.

l  Test Dataset: 10 video sequences in the resolution of 1080p will be used for evaluation. Each sequence contains 96 frames. All the sequences will be in YUV 4:2:0 color space. These video sequences will be distributed to all participates before a certain date. Participants are required to compress them within 72 hours.

 

        6. Timeline

l  March. 20, 2026, registration deadline for the competition. The authors can send the team’s name, team members, and the institution to xihua.sheng@polyu.edu.hk or lil1@ustc.edu.cn or cmjia@pku.edu.cn for registration

l  March. 25, 2026, release of the training and validation dataset

l  June. 13, 2026, submission of the decoder and docker environment (to the grand challenge organizers)

l  June. 13, 2026, release of the test dataset

l  June. 17, 2026, submission of the compressed bitstreams and decoded videos (to the grand challenge organizers)

l  June. 19, 2026, grand challenge paper submission due date (to the conference paper management system)

l  July. 17, 2026, winners, leader boards, and top creativity award notification.

l  July. 17, 2026, grand challenge paper acceptance notification.

l  July. 31, 2026, camera-ready grand challenge paper submission due date.

l  Sep. 22-24, 2026, challenge session at the MMSP 2026 conference. The grand challenge results, winners, and top creativity award will be announced. The top five participants will be invited to present at the conference.

        7. Organizers

l  Xihua Sheng, The Hong Kong Polytechnic University

l  Li Li, University of Science and Technology of China

l  Chuanmin Jia, Peking University

l  Zhu Li, University of Missouri-Kansas City

  For any inquiries, please email us at: xihua.sheng@polyu.edu.hk; lil1@ustc.edu.cn; cmjia@pku.edu.cn; lizhu@umkc.edu