Jinhua Zhang

I am a Ph.D. candidate at DIG at University of Electronic Science and Technology of China (UESTC), where I am advised by Prof. Shuhang Gu.

I received my bachelor's degree in Computer Science from Shandong University in 2023.

From 2023 to 2025, I worked as a research intern at Alibaba Cloud, mentored by Sijia Cai and Hualian Sheng.

Email / Google Scholar / Github

Research

My research interests lie in computer vision, with a focus on image generation and 3D generation.

	IDESplat: Iterative Depth Probability Estimation for Generalizable 3D Gaussian Splatting Wei Long, Haifeng Wu, Shiyin Jiang, Jinhua Zhang, Xinchun Ji, Shuhang Gu arXiv, 2026 code / arXiv IDESplat achieves SOTA feed-forward 3D reconstruction via iterative depth boosting with extreme parameter and memory efficiency.
	Texture Vector-Quantization and Reconstruction Aware Prediction for Generative Super-Resolution Qifan Li, Jiale Zou, Jinhua Zhang, Wei Long, Xingyu Zhou, Shuhang Gu ICLR, 2026 code / arXiv TVQ&RAP uses texture-specific VQ and reconstruction-aware prediction to slash quantization errors and improve super-resolution fidelity.
	MVAR: Visual Autoregressive Modeling with Scale and Spatial Markovian Conditioning Jinhua Zhang, Wei Long, Minghao Han, Weiyi You, Shuhang Gu ICLR, 2026 code / project page / arXiv MVAR redefines visual autoregression by applying Markovian constraints to scale and spatial, slashing training memory by 3.0x and enabling KV-cache-free inference without sacrificing performance.
	Generative Image Compression by Estimating Gradients of the Rate-variable Feature Distribution Minghao Han, Weiyi You, Jinhua Zhang, Leheng Zhang, Ce Zhu, Shuhang Gu arXiv, 2026 code / arXiv VCD reinterprets image compression as a stochastic forward diffusion path, enabling high-fidelity reconstruction via direct SDE reversal with minimal sampling steps.
	PerLDiff: Controllable Street View Synthesis Using Perspective-Layout Diffusion Model Jinhua Zhang, Hualian Sheng, Sijia Cai, Bing Deng, Qiao Liang, Wen Li, Ying Fu, Jieping Ye, Shuhang Gu ICCV, 2025 code / paper / arXiv PerLDiff leverages 3D geometric information within diffusion models to enable robust and precise object-level control for street view image generation.

Thank you to Jon Barron for the source code for the website!