He Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
He Wang has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Jinhui Tang
Nanjing Forestry University, Nanjing, China
6 shared publications - Longquan Dai
Nanjing University of Science and Technology, Nanjing, Jiangsu, China
6 shared publications - Shaomeng Wang
Nanjing University of Science and Technology, Nanjing, Jiangsu, China
5 shared publications - Dejiao Xue
Nanjing University of Science and Technology, Nanjing, Jiangsu, China
2 shared publications - Xiaolu Wei
Nanjing University of Science and Technology, Nanjing, Jiangsu, China
2 shared publications - Penghao Li
Nanjing University of Science and Technology, Nanjing, China
1 shared publication
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 10016-10025
HyperUC: A Flow Matching Hypernetwork for Unified Conditional Generation
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 10016-10025
Open AccessHyperUC: A Flow Matching Hypernetwork for Unified Conditional Generation
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12910-12919
ColorFlow: Robust Image Colorization with Pixel Flow Matching
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12910-12919
Open AccessColorFlow: Robust Image Colorization with Pixel Flow Matching
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 11381-11390
LAP: Controlling Text-to-Image Diffusion Models via Latent Alternation Projection
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 11381-11390
LAP: Controlling Text-to-Image Diffusion Models via Latent Alternation Projection
- Open Access
PC-Diffusion: Aligning Diffusion Models with Human Preferences via Preference Classifier
Diffusion models have achieved remarkable success in conditional image generation, yet their outputs often remain misaligned with human preferences. To address this, recent work has applied Direct Preference Optimization (DPO) to diffusion models, yielding significant improvements.~However, DPO-like methods exhibit two key limitations: 1) High computational cost,due to the entire model fine-tuning; 2) Sensitivity to reference model quality}, due to its tendency to introduce instability and bias. To overcome these limitations, we propose a novel framework for human preference alignment in diffusion models (PC-Diffusion), using a lightweight, trainable Preference Classifier that directly models the relative preference between samples. By restricting preference learning to this classifier, PC-Diffusion decouples preference alignment from the generative model, eliminating the need for entire model fine-tuning and reference model reliance.~We further provide theoretical guarantees for PC-Diffusion:1) PC-Diffusion ensures that the preference-guided distributions are consistently propagated across timesteps. 2)The training objective of the preference classifier is equivalent to DPO, but does not require a reference model.3) The proposed preference-guided correction can progressively steer generation toward preference-aligned regions.~Empirical results show that PC-Diffusion achieves comparable preference consistency to DPO while significantly reducing training costs and enabling efficient and stable preference-guided generation.
Open AccessPC-Diffusion: Aligning Diffusion Models with Human Preferences via Preference Classifier
Diffusion models have achieved remarkable success in conditional image generation, yet their outputs often remain misaligned with human preferences. To address this, recent work has applied Direct Preference Optimization (DPO) to diffusion models, yielding significant improvements.~However, DPO-like methods exhibit two key limitations: 1) High computational cost,due to the entire model fine-tuning; 2) Sensitivity to reference model quality}, due to its tendency to introduce instability and bias. To overcome these limitations, we propose a novel framework for human preference alignment in diffusion models (PC-Diffusion), using a lightweight, trainable Preference Classifier that directly models the relative preference between samples. By restricting preference learning to this classifier, PC-Diffusion decouples preference alignment from the generative model, eliminating the need for entire model fine-tuning and reference model reliance.~We further provide theoretical guarantees for PC-Diffusion:1) PC-Diffusion ensures that the preference-guided distributions are consistently propagated across timesteps. 2)The training objective of the preference classifier is equivalent to DPO, but does not require a reference model.3) The proposed preference-guided correction can progressively steer generation toward preference-aligned regions.~Empirical results show that PC-Diffusion achieves comparable preference consistency to DPO while significantly reducing training costs and enabling efficient and stable preference-guided generation.
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 10325-10333
Conducting Conditional Diffusion by Estimating the Mean Vector of von Mises-Fisher Distribution
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 10325-10333
Open AccessConducting Conditional Diffusion by Estimating the Mean Vector of von Mises-Fisher Distribution
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 10220-10228
Generative Semantic Probing for Vision-Language Models via Hierarchical Feature Optimization
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 10220-10228
Open AccessGenerative Semantic Probing for Vision-Language Models via Hierarchical Feature Optimization