📝 Publications

Towards Unified Human Motion-Language Understanding via Sparse Interpretable Characterization
Guangtao Lyu, Chenghao Xu, Jiexi Yan, Muli Yang, Cheng Deng

Project

We integrate the lexical representation paradigm into the motion-language representation framework, aligning both motion and text within a shared lexical vocabulary space. This integration significantly enhances interpretability and fosters a more intuitive and comprehensive understanding of human motion.

FETNet: Feature erasing and transferring network for scene text removal
Guangtao Lyu, Kun Liu, Anna Zhu, Seiichi Uchida, Brian Kenji Iwana

Project

• We propose a novel FETNet which could remove scene text near completely in images.

• Our method is formulated in a one stage way and is trained in an end to end manner.

• We introduce a novel Flickr ST dataset with multi category careful annotations.

PSSTRNet: Progressive Segmentation-Guided Scene Text Removal Network
Guangtao Lyu, Anna Zhu

Project

• We propose a novel STR network termed PSSTRNet. It decomposes the challenging STR task into two sim-ple subtasks and processes text segmentation and back-ground inpainting progressively.

• We design a Mask Update module and an adaptive fusion strategy to make full use of results from different iterations.

• Our proposed PSSTRNet is light-weighted and achieves SOTA quantitative and qualitative results on public synthetic and real scene datasets.