Tencent & Google Present 9 Papers at IEEE ICIP 2026: AV2 Video Coding & 3D Mesh Compression

Tencent and Google Joint Workshop at IEEE ICIP 2026

During the IEEE International Conference on Image Processing (ICIP) held September 13–17, 2026, Tencent Multimedia Lab and Google co‑hosted a Google‑Tencent workshop. The conference is the largest and most comprehensive event of the IEEE Signal Processing Society, focusing on image/video processing and computer vision.

Industry Roundtable on Video Coding Evolution

Tencent Distinguished Scientist Dr. Shan Liu participated in an industry roundtable alongside experts from Netflix (Anne Aaron), NVIDIA (Andrey Norkin), Nokia (Justin Ridge), and Huawei (Elena Alshina) to discuss the evolution and future of video coding technology.

Special Session on AV2 Coding Tools

Tencent expert researcher Dr. Liang Zhao and Google senior expert Debargha Mukherjee co‑chaired the session “Coding tools and features in the new AV2 video coding standard.” Dr. Zhao and Madhu P. Krishnan presented papers on intra prediction and transform techniques.

Poster Session: 3D Mesh Coding

Researchers Qingyang Zhou and Pranav Kadam showcased their latest work on 3D mesh coding in the poster session.

Nine Accepted Papers Overview

Tencent Multimedia Lab had nine papers accepted at ICIP 2026, covering multiple modules of the AV2 video compression standard and 3D mesh coding. The papers are summarized below.

01. Overview of intra prediction and intra mode coding in AV2

Authors: Liang Zhao, Tianqi Liu, Shan Liu, Jayasingam Adhuran, Qingyang Zhou, Jianle Chen, Cheng Chen, Raul Blazquez, Yixin Du, Mei Guo, Xin Zhao, Van Luong Pham, Mariana Afonso

Abstract: Intra prediction exploits spatial correlations within the same picture and is a core component of modern video compression systems. Compared to AV1, the AOMedia Video 2 (AV2) video coding format adopts a substantially enhanced intra prediction and intra mode coding framework to improve prediction accuracy and coding efficiency. This paper provides an overview of the intra prediction tools in AV2, including multiple reference line prediction, intra bi‑prediction, data‑driven intra prediction, implicit chroma‑from‑luma prediction, multi‑hypothesis cross‑component prediction, adaptive intra mode coding, and a 4‑tap interpolation filter, along with key modifications to the existing AV1 intra prediction tools. The motivation, design rationale, and coding impact of these tools are discussed to highlight how AV2 advances intra prediction beyond previous generations.

IEEE Xplore link:

https://ieeexplore.ieee.org/document/11630205

02. Overview of block partitioning framework in AV2

Authors: Urvang Joshi, Chi Yo Tsai, Yue Chen, Jayasingam Adhuran, Liang Zhao

Abstract: Block partitioning framework is a core component in any modern video coding standard, as it directly determines the block size used for predictions and transforms. Flexible block partitioning plays a crucial role in compression efficiency. This paper provides a technical overview of the block partitioning framework in AV2. Partitioning scheme for both coding blocks and transform blocks has been redesigned. Coding block partitioning is fully recursive with newly designed partitioning options. A newly introduced Semi‑Decoupled Partitioning (SDP) option provides additional flexibility by allowing luma and chroma components to have decoupled coding block partition trees. Transform block partitioning uses single‑level partitioning with more partitioning options. Tool‑off test results for several block partitioning aspects are provided.

IEEE Xplore abstract:

https://ieeexplore.ieee.org/abstract/document/11630267

03. An overview of inter coding tools in AV2

Authors: Yeqing Wu, Keng‑Shih Lu, Mohammed Golam Sarwer, Jianle Chen, Han Gao, Liang Zhao, Bohan Li, Guoxin Jin, Yunqing Wang, Lin Zheng, Jingning Han, Debargha Mukherjee, Hui Su, Yunfei Zheng, Alexis Tourapis, Shan Liu

Abstract: Inter prediction in AV2 is substantially enhanced over AV1 through new tools and improved Motion Vector (MV) signaling. Key additions include Temporal Interpolated Prediction (TIP), optical‑flow and sub‑block MV refinement, enhanced compound and weighted prediction, and new MV banks. The reference‑frame framework and Dynamic Reference List (DRL) are redesigned to reduce signaling while improving MV predictors with a simplified structure. Motion Vector Difference (MVD) coding is improved via adaptive and joint schemes, while Temporal Motion Vector Prediction (TMVP) is improved to increase coding efficiency and reduce memory footprint and bandwidth. Backward Reference Updating (BRU) enables low‑bandwidth, ultra‑low‑latency decoding for targeted applications. Performance gains achieved by these tools are demonstrated.

IEEE Xplore abstract:

https://ieeexplore.ieee.org/abstract/document/11630022

04. Overview of transform coding in AV2

Authors: Madhu P. Krishnan, Xin Zhao, Alican Nalci, Keng‑Shih Lu, Hilmi E. Egilmez, Aki Kuusela, Urvang Joshi, Van Luong Pham, Lin Zheng, Jingning Han, Kruthika Koratti Sivakumar

Abstract: This paper provides an overview and evaluation of transform coding tools in AV2, the successor to AV1 developed by the Alliance for Open Media. Key innovations include redesigned core transforms (DCT/DST), novel data‑driven transforms (DDT), intra and inter secondary transforms (IST), cross‑component transforms (CCTX), mode and block‑size dependent transform set (MBTS) selection for intra, DC‑based transform signaling improvements (DCTX), and large chroma transform. Collectively, these tools enable AV2 to deliver higher‑quality video at significantly reduced bitrates.

IEEE Xplore abstract:

https://ieeexplore.ieee.org/abstract/document/11630575

05. An overview of warp prediction in AV2

Authors: Mohammed Golam Sarwer, Yeqing Wu, Rachel Barker, Yunqing Wang, Debargha Mukherjee, Han Gao, Jayasingam Adhuran

Abstract: This paper summarizes warp prediction techniques developed in the AV2 video coding standard. To boost compression performance of warp modes, AV2 introduces several new tools: warp delta, warp extend, warp‑mv, warp‑new‑mv, 4x4 warp filter, and compound warp mode. Warp delta predicts a block's warp model from a Warp Reference List (WRL) and conditionally signals the delta between predicted and actual model. Warp extend generates a new warp parameter using both its motion vector and a neighboring block's warp parameters. Warp‑new‑mv predicts the translational component from the Dynamic Reference List (DRL) while non‑translational parameters come from the WRL. Simulation results show 4.18% and 5.60% bitrate reductions for random access and low delay configurations, respectively.

IEEE Xplore abstract:

https://ieeexplore.ieee.org/abstract/document/11630401

06. The in‑loop filtering pipeline in AV2

Authors: Debargha Mukherjee, Jianle Chen, Onur Guleryuz, Lin Zheng, In‑Suk Chong, Andrey Norkin, Yixin Du, Yunfei Zheng, Tianqi Liu, Khanh Quoc Dinh, Yangwoo Kim, Kwang Pyo Choi

Abstract: This paper presents an overview of the in‑loop filtering pipeline in the latest AV2 video codec from the Alliance for Open Media. Starting from AV1, the pipeline has been substantially enhanced by improving the deblocking filter, adding a cross‑component sample offset filter, replacing the filters in the Loop‑Restoration tool with new non‑separable and cross‑component Wiener filters, and adding a data‑driven guided detail filter. The paper provides detailed design of the tools and presents tool‑off results on a recent version of the AV2 reference software.

IEEE Xplore abstract:

https://ieeexplore.ieee.org/abstract/document/11630171

07. Residual and Entropy Coding in AV2

Authors: Alican Nalci, Hilmi E. Egilmez, Madhu P. Krishnan, Joe Young, Joel Sole, Xiaoqing Zhu, Zhijun Lei, Kruthika Koratti Sivakumar, Qingyang Zhou, Minhua Zhou

Abstract: This paper provides an overview of the residual and entropy coding tools adopted in AV2, including (i) coefficient coding within the adaptive transform coding (ATC) framework, (ii) probability adaptation rate adjustment (PARA) for improved probability estimation in entropy coding, (iii) adaptive truncated Rice coding for compression of large coefficient magnitudes, and (iv) forward skip coding (FSC) for coding transform‑skip residuals. These tools improve coding efficiency for both natural and screen content.

IEEE Xplore link:

https://ieeexplore.ieee.org/document/11630347

08. An overview of screen‑content coding in AV2

Authors: Qingyang Zhou, Van Luong Pham, Cheng Chen, Mohammed Golam Sarwer, Yingbin Wang, Aki Kuusela, Dzung Hoang, Guichun Li, Shan Liu

Abstract: Emerging applications like cloud gaming, remote desktop, and video conferencing stream computer‑generated content with sharp edges, repetitive patterns, and limited color variety. AV2 incorporates dedicated screen‑content coding (SCC) tools building upon previous designs. This paper analyzes and evaluates the main SCC tools in AV2, focusing on intra‑block copy (IntraBC) and palette‑based coding. Improvements include extended local search range for IntraBC, improved handling of global IntraBC with in‑loop filters, block‑adaptive weighted IntraBC prediction, sub‑pel block vector precision, advanced palette mode with line‑based coding and adaptive scanning order. Experimental results on representative screen content sequences show AV2 SCC tools achieve substantial bitrate reductions of 35% compared to non‑SCC coding configurations while maintaining comparable visual quality.

IEEE Xplore abstract:

https://ieeexplore.ieee.org/abstract/document/11630506

09. Rate distortion optimization for mesh Geometry compression

Authors: Qingyang Zhou, Pranav Kadam, Shan Liu, C.-C. Jay Kuo

Abstract: Rate control in mesh coding is usually achieved by varying quantization bit depths of geometric positions and other attributes, and by mesh simplification/remeshing. However, a formal rate‑distortion optimization (RDO) framework is absent in mesh codecs. This work introduces an RDO framework for mesh geometry coding, inspired by RDO in video codecs. It provides a way to select a prediction mode that minimizes the RD cost among candidate modes. The method is implemented in the AOMedia polygonal mesh codec (PMC) encoder with an average BD‑Rate gain of more than ‑3%. It can be seamlessly integrated into any mesh encoder that uses entropy coding to encode vertex positions.

ResearchGate publication:

https://www.researchgate.net/publication/413810362_Rate_Distortion_Optimization_for_Mesh_Geometry_Compression