Zhenning Shi*, Chen Xu*, Junhao Zhang, Kefei Zhang, Linjie Liu, Zhedong Zheng, Tao Li.
* Equal contribution.
Preprint: arXiv:2607.25275, 2026
ScaleResfusion introduces Residual Rectified Flow for adapting pretrained text-to-image rectified-flow models to real-world image restoration, combining residual transport, parameter-efficient fine-tuning, and knowledge distillation for efficient sampling.
Zhenning Shi, Haoshuai Zheng, Chen Xu, Changsheng Dong, Bin Pan, Xueshuo Xie, Along He, Tao Li, Huazhu Fu.
Published in NeurIPS, 2024
Start diffusion-based restoration process from noisy degraded images (not pure Gaussian noise) by injecting the residual term into the forward diffusion process, reducing unnecessary sampling.
Yang Yu, Chen Xu, Kai Wang.
Published in 2024 IEEE International Conference on Multimedia and Expo (ICME), 2024
This paper proposes TS-SAM (Two-Stream SAM), which enhances SAM with a lightweight convolutional side-branch adapter, a multi-scale refinement module, and a feature-fusion decoder, enabling it to achieve performance closer to, or even comparable with, specialized segmentation models on downstream segmentation tasks.
