Zhengqiang ZHANG (郑强 张)
I am currently a fourth-year PhD student in the Department of COMP at Hong Kong Polytechnic University (PolyU), supervised by Prof. Lei ZHANG. Prior to that, I worked as Multimedia AI Algorithm Engineer at HUYA Company for about one and a half years. I received my Master’s (supervised by Prof. Xinge You) and Bachelor’s degree at Huazhong University of Science and Technology (HUST). My research interests lie in image generation and image reconstruction. Recently, I focus on diffusion generative models and LLM-based agent frameworks, especially their native integration into image/video generation pipelines.
🔥 News
- 2026.09: 🎉🎉 2 papers accepted by NeurIPS 2026, including 1 first-author paper and 1 co-first-author paper.
- 2026.07: 🎉 1 paper accepted by ECCV 2026.
- 2026.02: 🎉 1 paper accepted by CVPR 2026.
- 2025.09: 🎉🎉 3 papers accepted by NeurIPS 2025, including 1 first-author paper.
📝 Publications
LDM-is-AE: Latent Diffusion Model is an Auto-Encoder for End-to-End Image Generation [paper] [code]
Zhengqiang ZHANG, Lingchen Sun, Rongyuan Wu, Qiaosi Yi, Xiangtao Kong, Chaodong Xiao, Lei Zhang+
NeurIPS 2026
+: Corresponding author
- A single-stage, end-to-end image generation framework that makes the DiT backbone itself an explicit auto-encoder, removing the external VAE and its inherited bias.
Memorize When Needed: Decoupled Memory Control for Spatially Consistent Long-Horizon Video Generation [paper] [code]
Yanjun Guo*, Zhengqiang ZHANG*, Pengfei Wang*, Xinyue Liang, Zhiyuan Ma, Lei Zhang+
NeurIPS 2026
*: Equal contribution, +: Corresponding author
- A decoupled framework that separates memory conditioning from generation for spatially consistent long-horizon video generation, cutting training costs while preserving the capacity for novel scene exploration.
GPSToken: Gaussian Parameterized Spatially-adaptive Tokenization for Image Representation and Generation [paper] [code]
Zhengqiang ZHANG, Rongyuan Wu, Lingchen Sun, Lei Zhang+
NeurIPS 2025
+: Corresponding author
- A non-uniform, spatially-adaptive image tokenizer using 2D Gaussians, which allocates more tokens for complex regions while less tokens for simple regions.
Generalized and Efficient 2D Gaussian Splatting for Arbitrary-scale Super-Resolution [paper] [code]
Du Chen*, Liyi Chen*, Zhengqiang ZHANG, Lei Zhang+
ICCV 2025
*: Equal contribution, +: Corresponding author
- Arbitrary-scale SR via 2D Gaussian Splatting.
FreCaS: Efficient Higher-Resolution Image Generation via Frequency-aware Cascaded Sampling [paper] [code]
Zhengqiang ZHANG, Ruihuang Li, Lei Zhang+
ICLR 2025
+: Corresponding author
- A training-free sampling strategy which generates higher-resolution images with pretrained diffusion models.
SSL: A Self-similarity Loss for Improving Generative Image Super-resolution [paper] [code]
Du Chen*, Zhengqiang ZHANG*, Jie Liang, Lei Zhang+
ACM MM 2024
*: Equal contribution, +: Corresponding author
- A self-similarity loss for generative ISR task.
TMP: Temporal Motion Propagation for Online Video Super-Resolution [paper] [code]
Zhengqiang ZHANG, Ruihuang Li, Shi Guo, Yang Cao, Lei Zhang+
IEEE TIP 2024
+: Corresponding author
- A fast motion estimation method for online VSR.
More papers can be seen in Google Scholar.
📖 Educations
- 2022.09 - Now, PhD, Hong Kong Polytechnic University, COMP Department.
- 2017.09 - 2020.06, Master, Huazhong University of Science and Technology, School of Electronic Information and Communications.
- 2013.09 - 2017.06, Bachelor, Huazhong University of Science and Technology, School of Electronic Information and Communications.
💻 Internships
- 2022.09 - Now, OPPO, China.
💼 Occupation
- 2020.09 - 2022.03, Huya, China.