We propose Bitstream Language Modeling as Robust Semantic Priors, a framework that learns bitstream-native semantic cues and injects them into vision models for video restoration, captioning, and human pose estimation under harsh bitstream corruption.
@article{huang2026bitstream,title={Towards Bitstream-corrupted Harsh Visual Understanding: Through Bitstream Language Modeling as Robust Semantic Priors},author={Huang, Chaoran and Li, Fangcheng and Liu, Tianyi and Liu, Wenyang and Wu, Kejun},journal={arXiv preprint arXiv:2608.21837},year={2026}}
arXiv
Bitstream Action Recognition is Byte Modeling
Fangcheng Li, Chaoran Huang, Tianyi Liu, and 5 more authors
We propose a byte-modeling framework for action recognition from corrupted bitstreams, with a dual-branch architecture, a real-world corruption simulator, and a benchmark spanning pixel, compressed, and bitstream domains.
@article{li2026bitstream,title={Bitstream Action Recognition is Byte Modeling},author={Li, Fangcheng and Huang, Chaoran and Liu, Tianyi and Liu, Wenyang and Wu, Kejun and Liu, Qiong and Yang, You and Li, Zhengguo},journal={arXiv preprint arXiv:2608.15695},year={2026}}
ACM MM
DriftAD: Visually-Guided Text Drift for Few-Shot Industrial Anomaly Detection
Wenyang Liu, Dongshuo Zhang, Tianyi Liu, and 2 more authors
CCF A
Proceedings of the ACM International Conference on Multimedia, 2026
We propose DriftAD, a visually-guided CLIP adaptation framework that adapts text representations to local visual context for few-shot industrial anomaly detection, achieving state-of-the-art results across 1-, 2-, and 4-shot settings with up to 1.5-point gains in AUROC and PRO.
@inproceedings{liu2026driftad,title={{DriftAD}: Visually-Guided Text Drift for Few-Shot Industrial Anomaly Detection},author={Liu, Wenyang and Zhang, Dongshuo and Liu, Tianyi and Wu, Kejun and Kong, Wai Kin (Adams)},booktitle={Proceedings of the ACM International Conference on Multimedia},year={2026}}
IEEE TMM
PromptSR: Cascade Prompting for Lightweight Image Super-Resolution
Wenyang Liu, Chen Cai, Jianjun Gao, and 4 more authors
We propose PromptSR, a cascade prompting framework for lightweight image super-resolution that leverages prompt-guided restoration to achieve efficient real-world image quality improvement.
@article{liu2026promptsr,title={{PromptSR}: Cascade Prompting for Lightweight Image Super-Resolution},author={Liu, Wenyang and Cai, Chen and Gao, Jianjun and Wu, Kejun and Wang, Yi and Yap, Kim-Hui and Chau, Lap-Pui},journal={IEEE Transactions on Multimedia},year={2026},publisher={IEEE}}
Information Fusion
LLMArk: Instance-Aware Foundation Model for Flood Risk Assessment
Kejun Wu, Shen Wang, Jianjun Gao, and 4 more authors
LLMArk introduces an instance-aware foundation model for comprehensive flood risk assessment using multimodal learning.
@article{wu2026llmark,title={{LLMArk}: Instance-Aware Foundation Model for Flood Risk Assessment},author={Wu, Kejun and Wang, Shen and Gao, Jianjun and Liu, Wenyang and Liu, Qiong and Cai, Chengtao and Yang, You},journal={Information Fusion},pages={104440},year={2026},publisher={Elsevier}}
PR
Corrupted Bitstream Semantic Understanding by Adaptive-Modal Large Language Models
Kejun Wu, Fangcheng Li, Wenyang Liu, and 2 more authors
We propose an adaptive-modal large language model approach for semantic understanding of corrupted bitstream data.
@article{wu2026corrupted,title={Corrupted Bitstream Semantic Understanding by Adaptive-Modal Large Language Models},author={Wu, Kejun and Li, Fangcheng and Liu, Wenyang and Liu, Qiong and Yang, You},journal={Pattern Recognition},volume={180},pages={114151},year={2026},publisher={Elsevier}}
JVCIR
SSH-Net: A Self-Supervised and Hybrid Network for Noisy Image Watermark Removal
Wenyang Liu, Jianjun Gao, and Kim-Hui Yap
CCF CCAS Q3JCR Q2
Journal of Visual Communication and Image Representation, 2025
SSH-Net is a self-supervised hybrid network for removing noisy watermarks from images without requiring paired clean/watermarked training data.
@article{liu2025ssh,title={{SSH-Net}: A Self-Supervised and Hybrid Network for Noisy Image Watermark Removal},author={Liu, Wenyang and Gao, Jianjun and Yap, Kim-Hui},journal={Journal of Visual Communication and Image Representation},pages={104516},year={2025},publisher={Academic Press}}
ACM MM
From Semantics, Scene to Instance-awareness: Distilling Foundation Model for Open-vocabulary Grounded Situation Recognition
Chen Cai, Tianyi Liu, Jianjun Gao, and 5 more authors
CCF A
Proceedings of the 33rd ACM International Conference on Multimedia, 2025
We distill foundation model knowledge across semantic, scene, and instance levels for open-vocabulary grounded situation recognition.
@inproceedings{cai2025semantics,title={From Semantics, Scene to Instance-awareness: Distilling Foundation Model for Open-vocabulary Grounded Situation Recognition},author={Cai, Chen and Liu, Tianyi and Gao, Jianjun and Liu, Wenyang and Wu, Kejun and Wang, Ruoyu and Wang, Yi and Liew, Soo Chin},booktitle={Proceedings of the 33rd ACM International Conference on Multimedia},pages={392--401},year={2025}}
KBS
CL-HOI: Cross-Level Human-Object Interaction Distillation from Multimodal Large Language Models
Jianjun Gao, Chen Cai, Ruoyu Wang, and 4 more authors
CL-HOI distills cross-level HOI knowledge from multimodal large language models for improved human-object interaction detection.
@article{gao2025cl,title={{CL-HOI}: Cross-Level Human-Object Interaction Distillation from Multimodal Large Language Models},author={Gao, Jianjun and Cai, Chen and Wang, Ruoyu and Liu, Wenyang and Yap, Kim-Hui and Garg, Kratika and Han, Boon Siew},journal={Knowledge-Based Systems},volume={320},pages={113561},year={2025},publisher={Elsevier}}
Information Fusion
MotionAnimate: Animate Human Images with Pose Motion for Vivid and Temporally Consistent Video Generation
Ruoyu Wang, Shaowei Wang, Rui Gong, and 5 more authors
MotionAnimate generates temporally consistent human videos by animating static images with pose motion sequences.
@article{wang2025motionanimate,title={{MotionAnimate}: Animate Human Images with Pose Motion for Vivid and Temporally Consistent Video Generation},author={Wang, Ruoyu and Wang, Shaowei and Gong, Rui and Cai, Chen and Gao, Jianjun and Wang, Wenqian and Liu, Wenyang and Yap, Kim-Hui},journal={Information Fusion},pages={103559},year={2025},publisher={Elsevier}}
arXiv
HSNet: Heterogeneous Subgraph Network for Single Image Super-Resolution
Qiongyang Hu, Wenyang Liu, Wenbin Zou, and 3 more authors
HSNet proposes a heterogeneous subgraph network architecture for single image super-resolution, capturing multi-scale structural dependencies.
@article{hu2025hsnet,title={{HSNet}: Heterogeneous Subgraph Network for Single Image Super-Resolution},author={Hu, Qiongyang and Liu, Wenyang and Zou, Wenbin and Su, Yuejiao and Chau, Lap-Pui and Wang, Yi},journal={arXiv preprint arXiv:2510.06564},year={2025}}
IEEE TMM
ByteNet: Rethinking Multimedia File Fragment Classification through Visual Perspectives
Wenyang Liu, Kejun Wu, Tianyi Liu, and 3 more authors
We propose ByteNet, which rethinks file fragment classification by converting byte sequences into visual representations, enabling CNN-based analysis for robust digital forensics.
@article{liu2024bytenet,title={{ByteNet}: Rethinking Multimedia File Fragment Classification through Visual Perspectives},author={Liu, Wenyang and Wu, Kejun and Liu, Tianyi and Wang, Yi and Yap, Kim-Hui and Chau, Lap-Pui},journal={IEEE Transactions on Multimedia},volume={27},pages={1305--1319},year={2024},publisher={IEEE}}
KBS
Intra-and Inter-Sector Contextual Information Fusion with Joint Self-Attention for File Fragment Classification
Yi Wang, Wenyang Liu, Kejun Wu, and 2 more authors
We propose a joint self-attention mechanism that fuses intra- and inter-sector contextual information for robust file fragment classification.
@article{wang2024intra,title={Intra-and Inter-Sector Contextual Information Fusion with Joint Self-Attention for File Fragment Classification},author={Wang, Yi and Liu, Wenyang and Wu, Kejun and Yap, Kim-Hui and Chau, Lap-Pui},journal={Knowledge-Based Systems},volume={291},pages={111565},year={2024},publisher={Elsevier}}
EMNLP
Empowering Large Language Model for Continual Video Question Answering with Collaborative Prompting
Chen Cai, Zheng Wang, Jianjun Gao, and 4 more authors
CCF B
Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 2024
We empower LLMs for continual video question answering through collaborative prompting, enabling adaptation to evolving video content without catastrophic forgetting.
@inproceedings{cai2024empowering,title={Empowering Large Language Model for Continual Video Question Answering with Collaborative Prompting},author={Cai, Chen and Wang, Zheng and Gao, Jianjun and Liu, Wenyang and Lu, Ye and Zhang, Runzhong and Yap, Kim-Hui},booktitle={Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing},year={2024}}
ICIP
CM^2-Net: Continual Cross-Modal Mapping Network for Driver Action Recognition
Ruoyu Wang, Chen Cai, Wenqian Wang, and 4 more authors
CCF C
IEEE International Conference on Image Processing, 2024
CM2-Net introduces a continual cross-modal mapping network for driver action recognition that adapts across modalities without forgetting previous knowledge.
@inproceedings{wang2024cm2,title={{CM$^2$-Net}: Continual Cross-Modal Mapping Network for Driver Action Recognition},author={Wang, Ruoyu and Cai, Chen and Wang, Wenqian and Gao, Jianjun and Lin, Dan and Liu, Wenyang and Yap, Kim-Hui},booktitle={IEEE International Conference on Image Processing},pages={2236--2242},year={2024},organization={IEEE}}
CVPR
Bitstream-Corrupted JPEG Images Are Restorable: Two-Stage Compensation and Alignment Framework for Image Restoration
Wenyang Liu, Yi Wang, Kim-Hui Yap, and 1 more author
CCF A
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023
We propose a two-stage compensation and alignment framework to restore images from corrupted JPEG bitstreams, recovering damaged visual content through compensation and spatial alignment strategies.
@inproceedings{liu2023bitstream_cvpr,title={Bitstream-Corrupted {JPEG} Images Are Restorable: Two-Stage Compensation and Alignment Framework for Image Restoration},author={Liu, Wenyang and Wang, Yi and Yap, Kim-Hui and Chau, Lap-Pui},booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},pages={9979--9988},year={2023}}
NeurIPS
Bitstream-Corrupted Video Recovery: A Novel Benchmark Dataset and Method
Tianyi Liu, Kejun Wu, Yi Wang, and 3 more authors
CCF A
Advances in Neural Information Processing Systems, 2023
We introduce the first benchmark dataset for bitstream-corrupted video recovery and propose a novel recovery method for damaged video streams.
@article{liu2023bitstream_neurips,title={Bitstream-Corrupted Video Recovery: A Novel Benchmark Dataset and Method},author={Liu, Tianyi and Wu, Kejun and Wang, Yi and Liu, Wenyang and Yap, Kim-Hui and Chau, Lap-Pui},journal={Advances in Neural Information Processing Systems},volume={36},pages={68420--68433},year={2023}}
AICAS
A Byte Sequence Is Worth an Image: CNN for File Fragment Classification Using Bit Shift and n-Gram Embeddings
Wenyang Liu, Yi Wang, Kejun Wu, and 2 more authors
IEEE International Conference on Artificial Intelligence Circuits and Systems, 2023
We demonstrate that byte sequences can be treated as images for file fragment classification, using bit shift and n-gram embeddings with CNN architectures.
@inproceedings{liu2023byte,title={A Byte Sequence Is Worth an Image: {CNN} for File Fragment Classification Using Bit Shift and n-Gram Embeddings},author={Liu, Wenyang and Wang, Yi and Wu, Kejun and Yap, Kim-Hui and Chau, Lap-Pui},booktitle={IEEE International Conference on Artificial Intelligence Circuits and Systems},year={2023}}
DCC
A Spatial-Focal Error Concealment Scheme for Corrupted Focal Stack Video
Kejun Wu, Yi Wang, Wenyang Liu, and 2 more authors
We propose a spatial-focal error concealment scheme for corrupted focal stack video, recovering visual content from damaged multi-focus video streams.
@inproceedings{wu2023spatial,title={A Spatial-Focal Error Concealment Scheme for Corrupted Focal Stack Video},author={Wu, Kejun and Wang, Yi and Liu, Wenyang and Yap, Kim-Hui and Chau, Lap-Pui},booktitle={Data Compression Conference},pages={91--100},year={2023},organization={IEEE}}
ISCAS
Image Representation and Deep Inception-Attention for File-Type and Malware Classification
Yi Wang, Kejun Wu, Wenyang Liu, and 2 more authors
CCF B
IEEE International Symposium on Circuits and Systems, 2023
We combine image representation with deep inception-attention networks for robust file-type and malware classification.
@inproceedings{wang2023image,title={Image Representation and Deep Inception-Attention for File-Type and Malware Classification},author={Wang, Yi and Wu, Kejun and Liu, Wenyang and Yap, Kim-Hui and Chau, Lap-Pui},booktitle={IEEE International Symposium on Circuits and Systems},pages={1--5},year={2023},organization={IEEE}}
ASP-DAC
MindReading: An Ultra-Low-Power Photonic Accelerator for EEG-Based Human Intention Recognition
Qian Lou*, Wenyang Liu*, Weichen Liu, and 2 more authors
* Co-first authors
CCF C
Asia and South Pacific Design Automation Conference, 2020
MindReading is an ultra-low-power photonic accelerator for EEG-based human intention recognition.
@inproceedings{lou2020mindreading,title={{MindReading}: An Ultra-Low-Power Photonic Accelerator for {EEG}-Based Human Intention Recognition},author={Lou, Qian and Liu, Wenyang and Liu, Weichen and Guo, Feng and Jiang, Lei},booktitle={Asia and South Pacific Design Automation Conference},pages={464--469},year={2020},organization={IEEE}}
DATE
HolyLight: A Nanophotonic Accelerator for Deep Learning in Data Centers
Weichen Liu, Wenyang Liu, Yichen Ye, and 3 more authors
Second author; first author was my master’s advisor
CCF B140+ citations
Design, Automation & Test in Europe Conference, 2019
HolyLight is a nanophotonic accelerator for efficient deep learning inference in data centers.
@inproceedings{liu2019holylight,title={{HolyLight}: A Nanophotonic Accelerator for Deep Learning in Data Centers},author={Liu, Weichen and Liu, Wenyang and Ye, Yichen and Lou, Qian and Xie, Yiyuan and Jiang, Lei},booktitle={Design, Automation \& Test in Europe Conference},pages={1483--1488},year={2019},organization={IEEE}}
ICPADS
Fine-Grained Task-Level Parallel and Low Power H.264 Decoding in Multi-Core Systems
Wenyang Liu, Weichen Liu, Mengquan Li, and 4 more authors
CCF C
IEEE 24th International Conference on Parallel and Distributed Systems, 2018
A fine-grained task-level parallel approach for low-power H.264 video decoding on multi-core systems.
@inproceedings{liu2018fine,title={Fine-Grained Task-Level Parallel and Low Power {H.264} Decoding in Multi-Core Systems},author={Liu, Wenyang and Liu, Weichen and Li, Mengquan and Chen, Peng and Yang, Lei and Xiao, Chunhua and Ye, Yaoyao},booktitle={IEEE 24th International Conference on Parallel and Distributed Systems},pages={307--314},year={2018},organization={IEEE}}