publications

2026

1. ECCV 2026

   

   Unison: Harmonizing Motion, Speech, and Sound for Human-Centric Audio-Video Generation

   Shihao Cheng, Jiaxu Zhang, Quanyue Song, and 6 more authors

   First Author

   Paper Bib Project Page

   We propose Unison, a unified framework for human-centric audio-video generation that harmonizes motion, speech, and ambient sound. Unison resolves the speech-SFX interference and motion-audio desynchronization problems that have long limited audio-video generative models, and produces temporally aligned, semantically consistent multi-modal outputs.

   @misc{cheng2026unisonharmonizingmotionspeech,
     title = {Unison: Harmonizing Motion, Speech, and Sound for Human-Centric Audio-Video Generation},
     author = {Cheng, Shihao and Zhang, Jiaxu and Song, Quanyue and Liu, Shansong and Guo, Zhizhi and Zhang, Xiaolei and Zhang, Chi and Li, Xuelong and Tu, Zhigang},
     year = {2026},
     eprint = {2605.08729},
     archiveprefix = {arXiv},
     primaryclass = {cs.CV},
     url = {https://arxiv.org/abs/2605.08729},
   }

2. ECCV 2026

   

   InteractiveAvatar: Real-Time Streaming Video Generation for Consistent and Intent-Aware Avatars

   Quanyue Song, Yishan He, Yanfei Zhang, and 6 more authors

   Paper Bib

   InteractiveAvatar tackles the understanding-interaction gap and the long-term semantic drift in infinite-length video generation. We design a long-short term token memory mechanism that lets the avatar follow user intent and remain visually consistent across arbitrarily long streams.

   @misc{song2026interactiveavatarrealtimestreamingvideo,
     title = {InteractiveAvatar: Real-Time Streaming Video Generation for Consistent and Intent-Aware Avatars},
     author = {Song, Quanyue and He, Yishan and Zhang, Yanfei and Cheng, Shihao and He, Zhixiang and Guo, Zhizhi and Zhang, Chi and Li, Xuelong and Jiang, Caigui},
     year = {2026},
     eprint = {2606.22905},
     archiveprefix = {arXiv},
     primaryclass = {cs.CV},
     url = {https://arxiv.org/abs/2606.22905},
   }

3. CVPR 2026

   

   GeoMMBench and GeoMMAgent: Toward Expert-Level Multimodal Intelligence in Geoscience and Remote Sensing

   Aoran Xiao, Shihao Cheng, Yonghao Xu, and 3 more authors

   Co-First Author

   In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Highlight , 2026

   Paper Bib Project Page Code Dataset

   We introduce GeoMMBench, the first expert-level multimodal evaluation benchmark for interdisciplinary, multi-sensor geoscience and remote-sensing scenarios, together with GeoMMAgent, a multi-agent architecture that achieves strong performance on the benchmark.

   @inproceedings{Xiao_2026_CVPR,
     author = {Xiao, Aoran and Cheng, Shihao and Xu, Yonghao and Ren, Yexian and Chen, Hongruixuan and Yokoya, Naoto},
     title = {GeoMMBench and GeoMMAgent: Toward Expert-Level Multimodal Intelligence in Geoscience and Remote Sensing},
     booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
     month = jun,
     year = {2026},
     pages = {34843-34853}
   }

4. T-CSVT 2025

   

   OwlSight: A Robust Illumination Adaptation Framework for Dark Video Human Action Recognition

   Shihao Cheng, Jinlu Zhang, Yue Liu, and 2 more authors

   First Author

   IEEE Transactions on Circuits and Systems for Video Technology, 2026

   Paper Bib

   OwlSight resolves the inefficient brightness utilization that bottlenecks low-light enhancement and action recognition end-to-end training. The framework jointly optimizes illumination adaptation and action recognition for robust dark-video understanding.

   @article{11250624,
     author = {Cheng, Shihao and Zhang, Jinlu and Liu, Yue and Xiao, Aoran and Tu, Zhigang},
     journal = {IEEE Transactions on Circuits and Systems for Video Technology},
     title = {OwlSight: A Robust Illumination Adaptation Framework for Dark Video Human Action Recognition},
     year = {2026},
     volume = {36},
     number = {4},
     pages = {4550-4564},
     keywords = {Videos;Lighting;Accuracy;Visualization;Training;Brightness;Random access memory;Learning systems;Complexity theory;Biomimetics;Dark video;action recognition;biomimetic design},
     doi = {10.1109/TCSVT.2025.3632359},
   }