Ref-GS: Directional Factorization for 2D Gaussian Splatting

Youjia Zhang, Anpei Chen †, Yumin Wan, Zikai Song, Junqing Yu, Yawei Luo, Wei Yang †
Computer Vision and Pattern Recognition (CVPR), 2025

Project | Code | Slides

• We propose Ref-GS, a method for directional light factorization in 2D Gaussian splatting, enabling photorealistic rendering and accurate geometry recovery. By applying directional encoding and spherical mip-grids, it reduces ambiguity and enhances roughness-aware shading. Our approach minimizes overhead and improves performance across open-world scenes.

Optimized View and Geometry Distillation from Multi-view Diffuser

Youjia Zhang, Zikai Song, Junqing Yu, Yawei Luo, Wei Yang †
International Joint Conference on Artificial Intelligence (IJCAI), 2025

Project | Code

• We propose USD, a method for consistent single-to-multi-view image synthesis and geometry recovery using image-conditioned diffusion and radiance fields. By treating the radiance field optimized during geometry extraction as a rigid multi-view consistency prior and introducing Unbiased Score Distillation (USD)—which leverages unconditioned noise from a 2D diffusion model to correct traditional optimization bias—we greatly enhance radiance field fidelity. Our two-step specialization of the 2D diffuser then performs object-specific denoising to generate high-quality multi-view images, from which we directly recover accurate geometry and texture. The result matches state-of-the-art performance on extensive datasets while preserving full freedom of camera positioning.

CA-Diff: Collaborative Anatomy Diffusion for Brain Tissue Segmentation

Qilong Xing, Zikai Song*, Yuteng Ye, Yuke Chen, Youjia Zhang, Na Feng, Junqing Yu, Wei Yang †
International Conference on Multimedia & Expo (ICME), 2025

Paper | Code

• We propose Collaborative Anatomy Diffusion (CA-Diff), a framework integrating spatial anatomical features to enhance segmentation accuracy of the diffusion model.

Dynamic feature pruning and consolidation for occluded person re-identification

YuTeng Ye, Jiale Cai, Chenxing Gao, Youjia Zhang, Junle Wang, Qiang Hu, Junqing Yu, Wei Yang †
AAAI Conference on Artificial Intelligence (AAAI), 2024

Paper | Code

• We propose a Feature Pruning and Consolidation (FPC) framework to circumvent explicit human structure parse, which consists of a sparse encoder, a global and local feature ranking module, and a feature consolidation decoder.

Progressive Text-to-Image Diffusion with Soft Latent Direction

YuTeng Ye, Jiale Cai, Hang Zhou, Guanwen Li, Youjia Zhang, Zikai Song, Chenxing Gao, Junqing Yu, Wei Yang †
AAAI Conference on Artificial Intelligence (AAAI), 2024

Paper | Code

• We propose to harness the capabilities of a Large Language Model (LLM) to decompose text descriptions into coherent directives adhering to stringent formats and progressively generate the target image.

AMD: anatomical motion diffusion with interpretable motion decomposition and fusion

Beibei Jing, Youjia Zhang, Zikai Song, Junqing Yu, Wei Yang †
AAAI Conference on Artificial Intelligence (AAAI), 2024

Paper

• We propose the Adaptable Motion Diffusion (AMD) model, which leverages a Large Language Model (LLM) to parse the input text into a sequence of concise and interpretable anatomical scripts that correspond to the target motion.

NeMF: Inverse volume rendering with neural microflake field

Youjia Zhang, Teng Xu, Junqing Yu, Yuteng Ye, Yanqing Jing, Junle Wang, Jingyi Yu, Wei Yang †
International Conference on Computer Vision (ICCV), 2023

Project | Code

• We propose to conduct inverse volume rendering by representing a scene using microflake volume, which assumes the space is filled with infinite small flakes and light reflects or scattersat each spatial location according to microflake distributions.

Highly accurate and large-scale collision cross sections prediction with graph neural networks

Renfeng Guo*, Youjia Zhang*, Yuxuan Liao*, Qiong Yang, Ting Xie, Xiaqiong Fan, Zhonglong Lin, Yi Chen, Hongmei Lu †, Zhimin Zhang †
Communications Chemistry (JCR Q1), 2023

Code

• We present SigmaCCS, a graph neural network-based method for CCS prediction from 3D conformers. It achieves high accuracy and chemical interpretability, enabling large-scale in-silico CCS estimation.