Hanyang Wang

Hanyang Wang | 汪晗阳

I'm currently a second-year master's student in the Department of Electronic Engineering at Tsinghua University , advised by Prof. Yueqi Duan. In 2025, I obtained my B.Eng. in the Department of Computer Science and Technology at Tsinghua University. I'm also working closely with Fangfu Liu.

My research interests lie in 3D Computer Vision, Video Generation, and AIGC.

Email / CV / Scholar / Github

Preprints

* indicates equal contribution. Some papers are highlighted.

Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoning

MirroS Team

Technical Report, 2026

[Paper] [Code] [Project Page] [Blog]

We introduce an agentic framework that discovers executable world representations for physical reasoning. Structured code models entities, states, events, and relations, allowing agents to simulate physical dynamics and verify their reasoning.

HarnessEval-W: Agentifying the Evaluation of Visual Worlds

MirroS Team

Technical Report, 2026

[arXiv] [Code] [Project Page] [Blog]

We introduce an agent-based evaluation framework for visual world models. Specialized agents solve measurable subproblems and organize their findings into an interpretable evidence tree, producing evaluations that closely align with human preferences.

Physics3D: Learning Physical Properties of 3D Gaussians via Video Diffusion

Fangfu Liu*, Hanyang Wang*, Shunyu Yao , Shengjun Zhang, Jie Zhou, Yueqi Duan

arXiv preprint, 2024

[arXiv] [Code] [Project Page]

In this paper, we propose Physics3D, a novel method for learning various physical properties of 3D objects through a video diffusion model. Our approach involves designing a highly generalizable physical simulation system based on a viscoelastic material model, which enables us to simulate a wide range of materials with high-fidelity capabilities.

Publications

	High-Fidelity Implicit Text-to-Shape Generation with Voxelized Diffusion Yueqi Duan, Fangfu Liu, Hanyang Wang, Muheng Li, Jiwen Lu, Jie Zhou International Journal of Computer Vision (IJCV), 2026 [Paper] We introduce Diffusion-SDF for text-conditioned 3D shape generation using voxelized signed distance fields. Diffusion-SDF++ further improves geometric detail through octree-based diffusion that adaptively allocates higher-resolution voxels to complex local structures.
	Learning Generalizable Semantic Radiance Fields with Cross-Reprojection Attention Yueqi Duan, Fangfu Liu, Jiawei Chi, Hanyang Wang, Chubin Zhang, Yu Zheng, Jie Zhou International Journal of Computer Vision (IJCV), 2026 [Paper] We introduce S-Ray and S-Frustum for generalizable semantic radiance fields. Cross-reprojection attention aggregates multi-view semantic information, while frustum-based rendering improves prediction consistency through efficient attention over a scene tensor.
	CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidance Hanyang Wang* , Yiyang Liu, Jiawei Chi, Fangfu Liu , Ran Xue , Yueqi Duan IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026 (Highlight)* [arXiv] [Code] [Project Page] We propose CFG-Ctrl, a unified framework that reinterprets CFG as a control applied to the first-order continuous-time generative flow. Based on this control-theoretic view, we further introduce SMC-CFG, which uses sliding mode control to drive the flow toward a rapidly convergent sliding manifold.
	ReconX: Reconstruct Any Scene from Sparse Views with Video Diffusion Model Fangfu Liu, Wenqiang Sun , Hanyang Wang* , Yikai Wang , Haowen Sun , Junliang Ye , Jun Zhang , Yueqi Duan IEEE Transactions on Image Processing (TIP), 2025 [arXiv] [Code] [Project Page] In this paper, we propose ReconX, a novel 3D scene reconstruction paradigm that reframes the ambiguous reconstruction challenge as a temporal generation task. The key insight is to unleash the strong generative prior of large pre-trained video diffusion models for sparse-view reconstruction.
	DreamReward-X: Boosting High-Quality 3D Generation with Human Preference Alignment Fangfu Liu , Junliang Ye , Yikai Wang , Hanyang Wang , Zhengyi Wang , Jun Zhu , Yueqi Duan IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2025 [Paper] [Code] [Project Page] We present a comprehensive framework, coined DreamReward-X, where we introduce a reward-aware noise sampling strategy to unleash text-driven diversity during the generation process while ensuring human preference alignment. Our results demonstrate the great potential for learning from human feedback to improve 3D generation.
	LangScene-X: Reconstruct Generalizable 3D Language-Embedded Scenes with TriMap Video Diffusion Fangfu Liu, Hao Li , Jiawei Chi , Hanyang Wang , Minghui Yang, Fudong Wang, Yueqi Duan IEEE International Conference on Computer Vision (ICCV), 2025 [arXiv] [Code] [Project Page] In this paper, we introduce a novel generative framework, coined LangScene-X, to unify and generate 3D consistent multi-modality information for reconstruction and understanding.
	Video-T1: Test-Time Scaling for Video Generation Fangfu Liu, Hanyang Wang , Yimo Cai, Kaiyan Zhang , Xiaohang Zhan , Yueqi Duan IEEE International Conference on Computer Vision (ICCV), 2025 [arXiv] [Code] [Project Page] We present the generative effects and performance improvements of video generation under test-time scaling (TTS) settings. The videos generated with TTS are of higher quality and more consistent with the prompt than those generated without TTS.
	VideoScene: Distilling Video Diffusion Model to Generate 3D Scenes in One Step Hanyang Wang, Fangfu Liu, Jiawei Chi , Yueqi Duan IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025 (Highlight) [arXiv] [Code] [Project Page] In this paper, we propose VideoScene to distill the video diffusion model to generate 3D scenes in one step, aiming to build an efficient and effective tool to bridge the gap from video to 3D.
	Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image Kailu Wu , Fangfu Liu, Zhihan Cai, Runjie Yan, Hanyang Wang, Yating Hu, Yueqi Duan , Kaisheng Ma Conference on Neural Information Processing Systems (NeurIPS), 2024 [arXiv] [Code] [Project Page] In this work, we introduce Unique3D, a novel image-to-3D framework for efficiently generating high-quality 3D meshes from single-view images, featuring state-of-the-art generation fidelity and strong generalizability. Unique3D can generate a high-fidelity textured mesh from a single orthogonal RGB image of any object in under 30 seconds.
	Make-Your-3D: Fast and Consistent Subject-Driven 3D Content Generation Fangfu Liu, Hanyang Wang, Weiliang Chen, Haowen Sun, Yueqi Duan European Conference on Computer Vision (ECCV), 2024 [arXiv] [Code] [Project Page] We introduce a novel 3D customization method, dubbed Make-Your-3D that can personalize high-fidelity and consistent 3D content from only a single image of a subject with text description within 5 minutes.

Honors and Awards

Outstanding Graduate, Department of Computer Science and Technology, Tsinghua University, 2025
Tsinghua Excellent Academic Scholarship, Tsinghua University, 2022

Website Template

Preprints

Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoning

HarnessEval-W: Agentifying the Evaluation of Visual Worlds

Physics3D: Learning Physical Properties of 3D Gaussians via Video Diffusion

Publications

High-Fidelity Implicit Text-to-Shape Generation with Voxelized Diffusion

Learning Generalizable Semantic Radiance Fields with Cross-Reprojection Attention

CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidance

ReconX: Reconstruct Any Scene from Sparse Views with Video Diffusion Model

DreamReward-X: Boosting High-Quality 3D Generation with Human Preference Alignment

LangScene-X: Reconstruct Generalizable 3D Language-Embedded Scenes with TriMap Video Diffusion

Video-T1: Test-Time Scaling for Video Generation

VideoScene: Distilling Video Diffusion Model to Generate 3D Scenes in One Step

Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image

Make-Your-3D: Fast and Consistent Subject-Driven 3D Content Generation

Honors and Awards