2025 Munich Tech Arena

Media Technology

group

Teams of
1 - 6 people

Video Compression

Pre- and post-processing for improving video compression

Access to detailed topic description, code resources and data

The objective of this task is to design a pair of pre-processing and post-processing filters to enhance the performance of an HEVC/h.265 [1] video codec, as illustrated in Fig. 1. Performance improvement is defined as either reducing the encoded video stream size while maintaining reconstruction quality, or enhancing reconstruction quality while keeping the stream size constant or a combination of both.

Fig.1: Block diagram of the video compression framework

To facilitate development, we provide a Python-based video compression framework, a test dataset, and benchmarking tools for performance evaluation. The framework includes placeholder modules for the pre- and post-processing filters, which participants are expected to replace with their own implementations.

In addition, the framework supports the transmission of metadata—arbitrary side information that can be embedded into the stream to aid post-processing. Importantly, the size of the metadata is counted toward the total stream size and must be considered when assessing overall performance.

Submissions will be evaluated based on the achieved BD-Rate [2], which reflects both the stream size and the quality of the decoded images.

Participants are free to adopt any approach, whether classical or learning-based. We encourage applicants to consult recent publications in top-tier conferences—such as CVPR and others for ideas and inspiration. According to recent studies [3], combining pre- and post-processing filters can lead to compression performance gains of 10% to 30%.

Provided input:

-Video Dataset: The dataset provided is in YUV [4] 4:2:0 [5] 10-bit format.

-Video encoder and decoder framework: The framework is based on the HEVC/h.265 standard, and includes Python example code for implementing custom pre- and post-processing filters.

-Test code: The provided test script includes example command lines that allow users to encode the dataset videos at various compression rates and evaluate compression performance.

Expected output:

-Pre- and post- processing algorithms: Participants must submit both the implementation code and a detailed description of their algorithms. In the case of learning-based solutions, participants are also required to provide the trained models along with sufficient materials to verify the training process, such as training code, configuration files, and etc.

-Encoded Streams and Reconstructed Frames: Participants must submit the compressed video streams generated by their solution as well as the corresponding reconstructed video frames.

Evaluation:

-BD-rate: Submissions will be evaluated using the BD-Rate metric, which reflects the trade-off between the size of the compressed video stream and the quality of the reconstructed frames across multiple quality points. Submissions demonstrating better rate-distortion performance will receive higher scores. Example code for computing the BD-Rate is provided as part of the framework.

-Objective image quality: The reconstruction quality will also be assessed using standard objective image quality metrics, including PSNR [6], MS-SSIM [7], and VMAF [8].

-Subjective image quality: In addition to objective metrics, subjective visual quality will be evaluated by expert judges to ensure that the results are not unreasonably overfitted to specific metrics.

-The complexity of the submission will also be considered for the final grade, however with a lower importance than reconstruction quality and stream size. The reason to consider complexity is to limit excessively complex and impractical solutions.

Additional information:

The framework utilizes the HEVC/h.265 codec from the FFmpeg library to compress video frames. Participants are not allowed to modify the codec itself; only the pre-processing and post-processing modules may be altered. To assist participants in getting started, the following example strategies may serve as inspiration:

-Color Distribution Adjustment: Adjust the color distribution or reduce saturation before compression, and restore the original color characteristics after decoding.

-Resolution Scaling: Downscale selected frames prior to compression and upscale them during post-processing to reduce bitrates while maintaining perceived visual quality [9].

-Noise Management: Denoise the input frames before encoding to reduce bitrate, and reintroduce synthetic or estimated noise patterns after decoding to preserve natural appearance [10].

-Intensity Transformation: Apply gamma correction or other non-linear intensity transformations before compression, and reverse them post-decoding to better preserve detail in specific luminance ranges [11].

Reference:

[1] High Efficiency Video Coding, https://en.wikipedia.org/wiki/High_Efficiency_Video_Coding

[2] Bjøntegaard Delta (BD): A Tutorial Overview of the Metric, Evolution, Challenges, and Recommendations. http://dx.doi.org/10.13140/RG.2.2.18622.05444

[3] Perceptual Video Compression with Neural Wrapping, https://cvpr.thecvf.com/virtual/2025/poster/34427

[4] YUV, https://en.wikipedia.org/wiki/Y%E2%80%B2UV

[5] Chroma Subsampling, https://en.wikipedia.org/wiki/Chroma_subsampling

[6] Peak signal-to-noise ratio, https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio

[7] Wang, Z.; Simoncelli, E.P.; Bovik, A.C. (2003-11-01). "Multiscale structural similarity for image quality assessment". The Thirty-Seventh Asilomar Conference on Signals, Systems & Computers, 2003. Vol. 2. pp. 1398–1402 Vol.2. https://doi.org/10.1109/ACSSC.2003.1292216

[8] Video Multimethod Assessment Fusion https://en.wikipedia.org/wiki/Video_Multimethod_Assessment_Fusion

[9] Adaptive Resolution Change for Versatile Video Coding, https://ieeexplore.ieee.org/document/9301762

[10] AOMedia Film Grain Synthesis 1, https://norkin.org/research/afgs1/

[11] Luma Mapping with Chroma Scalling in Versatile Video Coding, https://sigport.org/sites/default/files/docs/DCC2020_ID189_LMCS_present.pdf

Two-Part Evaluation Structure

The competition consists of two evaluation parts:

Part 1 – Idea Submission Evaluation
All teams are required to submit a written proposal outlining their solution approach by September 15, 2025.
👉 [Click here to view the idea template]
Submissions will be reviewed by our judging panel to ensure they meet the template and content requirements.
If your submission is approved, your team will receive full marks for Part 1, which accounts for 20% of the final score.

Part 2 – Final Project Evaluation

All teams are required to submit a final project approach by October 15, 2025.

The final deliverable will be evaluated by Huawei experts based on technical quality, innovation, and completeness.
This part contributes 80% of the total score.

Participate