Visual Prompting for One-shot Controllable Video Editing without Inversion
Zhengbo Zhang, Yuxi Zhou, Duo Peng, Joo Hwee Lim, Zhigang Tu, De Wen Soh, Lin Geng Foo
CVPR, 2025
paper

We perform one-shot controllable video editing via visual prompting to eschew the DDIM inversion process, which can potentially introdude errors.

Diff-Tracker: Text-to-Image Diffusion Models are Unsupervised Trackers
Zhengbo Zhang, Li Xu, Duo Peng, Hossein Rahmani, Jun Liu
ECCV, 2024
paper / code

We introduce Diff-Tracker, which leverages the rich knowledge encapsulated within the pre-trained diffusion model, such as the understanding of image semantics and structural information, to address unsupervised visual tracking.

Harnessing Text-to-Image Diffusion Models for Category-Agnostic Pose Estimation
Duo Peng, Zhengbo Zhang, Ping Hu, Qiuhong Ke, David Yau, Jun Liu
ECCV (Oral), 2024
paper

We propose to harness rich knowledge in the off-the-shelf text-to-image diffusion model to effectively address Category-Agnostic Pose Estimation, without training on carefully prepared base categories.

Distilling Inter-Class Distance for Semantic Segmentation
Zhengbo Zhang, Chunluan Zhou, Zhigang Tu
IJCAI (Long Oral), 2022
paper

We propose a novel knowledge distillation method for semantic segmentation that encourages the student model to achieve large inter-class distances in the feature space, thereby enhancing segmentation accuracy.

Instance Temperature Knowledge Distillation
Zhengbo Zhang, Yuxi Zhou, Jia Gong, Jun Liu, Zhigang Tu,
ArXiv Preprint, 2023
project page / paper / code

We formulate the allocation of instance-specific temperatures in knowledge distillation as a sequential decision-making task and propose a novel reinforcement learning-based method, RLKD, to address it.