Self-Rewarding Sequential Monte Carlo for Masked Diffusion Language Models
Ziwei Luo, Ziqi Jin, Lei Wang, Lidong Bing, Thomas B. Schön

Preprint | Paper | Project | Code

• This work introduces an inference-time scaling method that leverages trajectory-level confidence from diffusion models as importance weights to steer generation toward globally confident, high-quality samples.
• Self-Rewarding SMC is reward-free and thus can be applied to arbitrary pretrained models and tasks.

Forward-only Diffusion Probabilistic Models
Ziwei Luo, Fredrik K. Gustafsson, Jens Sjölund, Thomas B. Schön

Preprint | Paper | Project | Code

• This work introduces the mean reversion term into both the drift and diffusion functions, enabling high-quality data samples with a single diffusion process.
• A stochastic extension of flow matching.

Taming diffusion models for image restoration: a review
Ziwei Luo, Fredrik K. Gustafsson, Zheng Zhao, Jens Sjölund, Thomas B. Schön

Philosophical Transactions A 2025 | Paper

• This work summarizes key constructions in diffusions and survey contemporary techniques that make use of diffusion models in solving general image restoration tasks.

Entropy-regularized Diffusion Policy with Q-Ensembles for Offline Reinforcement Learning
Ruoqi Zhang$^\dagger$, Ziwei Luo$^\dagger$, Jens Sjölund, Thomas B. Schön, Per Mattsson

NeurIPS 2024 | Paper | Code

• This work presents a diffusion-based policy for offline reinforcement learning.
• A diffusion entropy and the Q-ensembles are introduced to improve the RL performance.

Controlling Vision-Language Models for Multi-Task Image Restoration
Ziwei Luo, Fredrik K. Gustafsson, Zheng Zhao, Jens Sjölund, Thomas B. Schön

ICLR 2024 | Paper | Project | Code

• This work presents a degradation-aware vision-language model (DA-CLIP) for multi-task image restoration.
• A controller predicts degradation types and also adapts the fixed CLIP image encoder to predict high-quality feature embeddings.

Image Restoration with Mean-Reverting Stochastic Differential Equations
Ziwei Luo, Fredrik K. Gustafsson, Zheng Zhao, Jens Sjölund, Thomas B. Schön

ICML 2023 | Paper | Project | Code

• This work presents a stochastic differential equation (SDE) approach for general-purpose image restoration.
• A maximum likelihood objective is proposed to learn an optimal reverse trajectory which stabilizes the training and improves accuracy.

Refusion: Enabling Large-Size Realistic Image Restoration with Latent-Space Diffusion Models
Ziwei Luo, Fredrik K. Gustafsson, Zheng Zhao, Jens Sjölund, Thomas B. Schön

CVPRW 2023 | Paper | Code

• Winning solution of the NTIRE Image Shadow Removal Challenge 2023.
• An extention of the IR-SDE.

BSRT: Improving Burst Super-Resolution with Swin Transformer and Flow-Guided Deformable Alignment
Ziwei Luo, Youwei Li, Shen Cheng, Lei Yu, Qi Wu, Zhihong Wen, Haoqiang Fan, Jian Sun, Shuaicheng Liu

CVPRW 2022 | Paper | Code

• Champion of the NTIRE Burst Super-Resolution Challenge 2022 in Real track.

Deep Constrained Least Squares for Blind Image Super-Resolution
Ziwei Luo, Haibin Huang, Lei Yu, Youwei Li, Haoqiang Fan, Shuaicheng Liu

CVPR 2022 | Paper | Code

• This work proposes to disentangle deblurring and upsampling in blind super-resolution, and provides a theoretical guidance to make use of the kernel.

Stochastic Planner-Actor-Critic for Unsupervised Deformable Image Registration
Ziwei Luo*, Jing Hu*, Xin Wang, Shu Hu, Bin Kong, Youbing Yin, Qi Song, Xi Wu, Siwei Lyu

AAAI 2022 | Paper | Code

• It is the first work that performs deformable image registration via high-dimensional action deep reinforcement learning.

EBSR: Feature enhanced burst super-resolution with deformable alignment
Ziwei Luo, Lei Yu, Xuan Mo, Youwei Li, Lanpeng Jia, Haoqiang Fan, Jian Sun, Shuaicheng Liu

CVPRW 2021 | Paper | Code

• Champion of the NTIRE Burst Super-Resolution Challenge 2021 in Real track