Recent Publications

Yun Zhang, Lewen Fan, Mao Cui, Xiaoxia Huang, Sam Kwong. IEEE Transactions on Circuits and Systems for Video Technology, 2026.

Summary: To account for the content-dependent perceptual importance of geometry and attributes in V-PCC, the work introduces PCQA-AVIGA and derives 2D/3D perceptual distortion and rate models. Two PBAO schemes then adapt geometry and attribute quantization under a total-rate constraint, reducing BDBR by 6.44% on average while improving perceptual quality.

Yun Zhang, Lewen Fan, Zixi Guo, Xu Wang, Xiaoxia Huang, Sam Kwong. IEEE Transactions on Image Processing, vol. 35, pp. 3451-3465, 2026.

Summary: RR-DPCC addresses the need to train separate models for different target rates by introducing a rate-reconfigurable codec with online and offline perceptual bit allocation. A single model supports flexible rate control while balancing geometry and attribute coding costs.

Yun Zhang, Zhiwei Guo, Zixi Guo, Linwei Zhu, C.-C. Jay Kuo. IEEE Transactions on Circuits and Systems for Video Technology, 2026.

Summary: Deep-JGAC provides end-to-end joint geometry and attribute compression for dense colored point clouds. Residual self-attention geometry coding, re-colorization, and joint entropy modeling improve the coordinated representation and coding efficiency of both components.

Junqi Liu, Yun Zhang, Xiaoqi Wang, Long Xu, Sam Kwong. IEEE Transactions on Multimedia, vol. 28, pp. 114-127, 2026.

Summary: The work defines JRD as the smallest difference recognizable by machine vision and formulates its prediction as multi-class classification. A Transformer jointly models content and distortion so the codec can remove unnecessary bits while preserving downstream analysis accuracy.

Yinan Chen, Yang Zhou, Xiaoxia Huang, Yun Zhang, Yuguang Fang. IEEE Transactions on Networking, vol. 34, pp. 4959-4973, 2026.

Summary: CausalFi addresses latent confounding and domain shifts in cross-domain Wi-Fi CSI-based human activity recognition. Its structural causal model combines causal feature selection, importance sampling, and counterfactual style augmentation to isolate stable action-related features, achieving 92.6% average accuracy across realistic domains and commercial Wi-Fi devices.

Renyi Chen, Liqian Zhang, Yun Zhang, Li Li, Fu Lee Wang, Zhaoqing Pan. IEEE Transactions on Multimedia, 2026.

Summary: CA-LIC differentiates luma and chroma coding according to their asymmetric importance to downstream machine-vision tasks. CASC reduces chroma volume through superpixel-level semantic aggregation and restores it with luma-guided compensation, while CFTM suppresses cross-component redundancy; the framework improves rate-task performance in object detection, instance segmentation, and semantic segmentation.

Zhiwei Guo, Yun Zhang, Linwei Zhu. IEEE Transactions on Broadcasting, vol. 72, no. 3, pp. 877-889, 2026.

Summary: DD-PCAC unifies motion estimation, compensation, and residual coding in an end-to-end differentiable framework for dynamic point cloud attributes. DMME combines dynamic ball matching with pixel- and feature-domain aggregation, while ST-CRC exploits spatiotemporal conditions for residual coding, reducing average bit rate by 31.57% and 30.33% over TSC-PCAC and G-PCC, respectively.

Xiaoqi Wang, Yun Zhang. ACM Transactions on Multimedia Computing, Communications and Applications, 2026.

Summary: GlintIQA combines ViT-derived global context with CNN-based fine-grained local distortion features and progressively fuses them through channel self-attention and spatial interaction. Its SAQT strategy transfers subjective scores between semantically similar images with identical distortions to build a large-scale training set, substantially improving cross-dataset generalization.

Yun Zhang, Feifan Chen, Na Li, Xu Wang, Fen Miao, Sam Kwong. IEEE Transactions on Image Processing, vol. 35, pp. 1305-1320, 2026.

Summary: JGAU jointly upsamples point cloud geometry and attributes by exploiting their spatial correlation. The study also establishes a large-scale colored point cloud upsampling dataset to support training and evaluation at multiple enlargement factors.

Shiyu Feng, Linwei Zhu, Yun Zhang, Na Li, Wenhui Wu, Shiqi Wang. IEEE Transactions on Multimedia, vol. 28, pp. 1756-1766, 2026.

Summary: C-CTX organizes latent context jointly across spatial and channel dimensions for learned image compression. Channel re-arrangement and multiple context prediction modes provide more accurate probability estimates and improve rate-distortion performance.

Yun Zhang, Shisheng Zhang, Na Li, Chunling Fan, Raouf Hamzaoui. IEEE Transactions on Circuits and Systems for Video Technology, vol. 36, no. 1, pp. 622-636, 2026.

Summary: VP-JND predicts the picture-wise distortion threshold that is just noticeable to human observers. The predicted threshold can guide coding parameters and bit allocation, reducing perceptual redundancy while maintaining visual quality.

Yun Zhang, Cui Mao, Na Li, Chunling Fan, Weisi Lin. IEEE Transactions on Multimedia, vol. 28, pp. 1894-1908, 2026.

Summary: The method regularizes colored point clouds into 2D geometry and attribute maps before extracting complementary cues. ViT-based geometry modeling, CNN-based attribute modeling, and quality regression are combined for objective colored point cloud assessment.

Linwei Zhu, Ruxu Liang, Yun Zhang, Hui Yuan, Sam Kwong. IEEE Transactions on Multimedia, 2026.

Summary: A 3D spatiotemporal search and temporal feature fusion pipeline addresses color noise and flicker in dynamic point clouds. Local feature extraction and Conv-PointLSTM exploit inter-frame consistency to recover more stable color attributes.

Renwei Tu, Gangyi Jiang, Zhongjie Zhu, Yeyao Chen, Ting Luo, Yun Zhang, Mei Yu. IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 10, no. 1, 2025.

Summary: A hierarchical restoration method targets structural artifacts caused by very-low-bitrate G-PCC octree coding. Distorted regions are located and corrected at multiple scales to improve geometry and visual quality under sparse coding conditions.

Huan Zhang, Xu Zhang, Linwei Zhu, Yun Zhang, Jiangzhong Cao, Wing-Kuen Ling. Expert Systems With Applications, vol. 264, art. no. 125853, 2025.

Summary: A perceptually guided restoration network jointly addresses compression artifacts and geometric warping in synthesized 3D views. Distortion-map prediction and restoration branches cooperate to repair regions that most strongly affect the viewing experience.

Huan Zhang, Xu Zhang, Nian Cai, Jianglei Di, Yun Zhang. Computer Vision and Image Understanding, vol. 261, art. no. 104491, 2025.

Summary: MDDA-former combines CNNs and Transformers with multi-dimensional dynamic attention to model local texture and long-range dependencies. Its unified encoder-decoder supports general restoration tasks such as denoising, deraining, and deblurring.

Shiyu Feng, Yun Zhang, Linwei Zhu, Sam Kwong. IEEE Transactions on Broadcasting, vol. 71, no. 3, pp. 889-902, 2025.

Summary: LFIC-DRASC targets the high-dimensional structure and cross-view redundancy of light field imagery. Disentangled representations, asymmetrical strip convolution, and a dedicated entropy model improve compression while preserving view consistency.

Xiaoqi Wang, Yun Zhang, Junqi Liu. Neurocomputing, vol. 650, art. no. 130731, 2025.

Summary: Multi-granular auxiliary embeddings are constructed along spatial and channel dimensions to strengthen perceptual quality features. A search strategy selects effective fusion positions and orders, improving generalization across datasets and distortion types.

Junle Liu, Yun Zhang, Zixi Guo, Xiaoxia Huang, Gangyi Jiang. ACM Transactions on Multimedia Computing, Communications, and Applications, vol. 21, no. 9, art. no. 263, 2025.

Summary: MFIBA allocates bits for multi-scale machine features according to their estimated importance. An online task loss-rate model balances feature coding cost against downstream task performance at each scale.

Lewen Fan, Yun Zhang. Journal of Visual Communication and Image Representation, vol. 113, art. no. 104610, 2025.

Summary: A texture-aware fast mode decision and complexity allocation framework reduces the cost of VVC-based point cloud video coding. Texture and coding-unit correlations prune unlikely modes early and dynamically distribute computation according to content.

Zixi Guo, Yun Zhang, Linwei Zhu, Hanli Wang, Gangyi Jiang. IEEE Transactions on Broadcasting, vol. 71, no. 1, pp. 154-166, 2025.

Summary: TSC-PCAC combines voxel Transformers and sparse convolutions for point cloud attribute compression. Its two-stage modules capture local neighborhoods and long-range dependencies, while a channel context model improves entropy coding of color attributes.

Linwei Zhu, Ruxu Liang, Yun Zhang, Gangyi Jiang, Yo-Sung Ho. IEEE Signal Processing Letters, vol. 32, pp. 2947-2951, 2025.

Summary: Geometry conditions a latent diffusion model for point cloud color-attribute denoising. Realistic compression-noise simulation and dynamic sampling-step control preserve structure while balancing restoration accuracy and computational efficiency.

Huan Zhang, Zhijun Xiong, Xu Zhang, Jiangzhong Cao, Yun Zhang. Digital Signal Processing, vol. 165, art. no. 105293, 2025.

Summary: ResNet-50 and the dual-tree complex wavelet transform extract multi-scale, multi-directional texture and structure cues. Global object-shift compensation, directional distortion statistics, and weighted regression provide targeted assessment of DIBR-synthesized images.

Yuyu Xu, Pingping Zhang, Minghui Chen, Zhuo Chen, Wenhui Wu, Yun Zhang, Xu Wang. ACM Transactions on Multimedia Computing, Communications, and Applications, vol. 21, no. 2, art. no. 58, 2025.

Summary: The RGB-D codec establishes bi-directional cross-modal connections in the encoder, decoder, and entropy model. Bi-CPT and Bi-CEE exchange color and depth priors and use joint training to improve rate-distortion performance for both modalities.

Jiali Zhang, Qinglan Li, Liguang Wu, Qifeng Qian, Xuyang Ge, Sam Kwong, Yun Zhang, Xinyan Lyu, Guanbo Zhou, Gaozhen Nie, Pak Wai Chan, Wai Kin Wong, Linwei Zhu. Weather and Climate Extremes, vol. 48, art. no. 100770, 2025.

Summary: The study links tropical cyclone inner-core symmetry with rapid intensification and builds multiple tree-based predictors plus an ensemble model. Satellite imagery, intensity, and environmental variables are fused to support objective forecasts at different lead times.

Mao Cui, Yun Zhang, Chunling Fan, Raouf Hamzaoui, Qinglan Li. IEEE Transactions on Multimedia, vol. 26, pp. 11111-11125, 2024.

Summary: CF-PCQA treats local structure in 3D point space and texture in 2D projections as complementary evidence. Multi-branch feature extraction and regression jointly learn how geometry, color, and spatial distribution affect perceived point cloud quality.

Yun Zhang, Haoqing Lin, Jing Sun, Linwei Zhu, Sam Kwong. IEEE Transactions on Multimedia, vol. 26, pp. 5925-5938, 2024.

Summary: The work refines just recognizable distortion from the image level to individual objects and constructs a corresponding data and prediction pipeline. Region-specific thresholds provide finer perceptual guidance for content-adaptive image and video compression.

Linwei Zhu, Yun Zhang, Na Li, Wenhui Wu, Shiqi Wang, Sam Kwong. IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 11, pp. 12092-12096, 2024.

Summary: Three neural filters are inserted at reference-pixel, coding-unit, and frame levels inside the VVC loop. They address local prediction error, block distortion, and global artifacts respectively, and their combination further improves reconstruction and coding efficiency.

Huan Zhang, Dongsheng Zheng, Yun Zhang, Jianzhong Cao, Weisi Lin, Wing-Kuen Ling. IEEE Transactions on Multimedia, vol. 26, pp. 6834-6847, 2024.

Summary: Wavelet decomposition and gradient-magnitude similarity capture black holes, stretching, and structural displacement in DIBR-synthesized views. Reference-view compensation and multi-scale fusion increase sensitivity to localized geometric distortion.

Chenghua Li, Zongze Li, Jing Sun, Yun Zhang, Xiaoping Jiang, Fan Zhang. ACM Transactions on Multimedia Computing, Communications, and Applications, vol. 20, no. 1, art. no. 12, 2024.

Summary: DGRNet applies dynamically weighted gradient reversal to visible-infrared person re-identification. Identity classification and modality discrimination are trained jointly, with adaptive adversarial strength reducing the modality gap while preserving identity cues.

Taowen Xu, Gangyi Jiang, Mei Yu, Yun Zhang, Zhidi Jiang, Yo-Sung Ho. IEEE Transactions on Emerging Topics in Computational Intelligence, vol. 8, no. 1, pp. 892-902, 2024.

Summary: A multi-view projection framework jointly repairs geometry holes and missing colors introduced by G-PCC Trisoup coding. Geometry-weighted repair and a gated color network work together before inverse projection reconstructs a more complete colored point cloud.

Chunling Fan, Yun Zhang, Linwei Zhu, Xinju Wu. IET Electronics Letters, vol. 60, no. 5, pp. 1-4, 2024.

Summary: A head-mounted AR subjective study builds a dataset containing 10 reference point clouds and 90 distorted versions. It analyzes how geometry and texture quantization affect six-degree-of-freedom viewing and provides a benchmark for objective quality metrics.

Wei Liu, Mei Yu, Zhidi Jiang, Haiyong Xu, Zhoujie Zhu, Yun Zhang, Gangyi Jiang. Digital Signal Processing, vol. 149, pp. 1-13, 2024.

Summary: Perceptual geometry coding targets cracks introduced during V-PCC projection, filling, and quantization. Improved geometry-map generation and coding control preserve continuity in visually important regions of reconstructed dynamic point clouds.

Hui Yuan, Raouf Hamzaoui, Ferrante Neri, Shengxiang Yang, Xin Lu, Linwei Zhu, Yun Zhang. Frontiers in Signal Processing, vol. 4, art. no. 1385287, 2024.

Summary: Rate-distortion optimization and differential evolution select V-PCC quantization parameters for groups of frames. Content- and rate-aware parameter search reduces bitrate while maintaining reconstructed point cloud quality.

Yun Zhang, Raouf Hamzaoui, Xu Wang, Junhui Hou, Giuseppe Valenzise. IET Electronics Letters, vol. 60, no. 14, pp. 1-2, 2024.

Summary: This guest editorial reviews recent deep-learning advances in point cloud processing, compression, and analysis. It covers sampling, enhancement, coding, quality assessment, and 3D understanding while summarizing the main contributions of the special issue.