Yichen Sheng

I am a research scientist in NVIDIA Applied Deep Learning Research(ADLR) team, working on the next-generation graphics.

My mission is to create a photorealistic virtual world (the ultimate goal of computer graphics, recent fancy name: world model) no matter in which form. Currently I focus on generative rendering and scalable ways to create a digital twin for real world. I believe this is the first step towards real spaital intelligence.

I am driven by the conviction that the most meaningful advances arise from deep insight rather than complexity. My work embodies this view by turning principled understanding into simple designs with enduring real-world impact.

Email / GitHub / Google Scholar / LinkedIn / CV

Industry Product

DLSS 5: Generative Neural Rendering

Core Contributor (Research, Data, Training & Infrastructure)
NVIDIA Applied Deep Learning Research
> project > paper > GTC video > NV Media > SIGGRAPH talk

Real-time generative rendering for interactive graphics.

Research

	I-Scene: 3D Instance Models are Implicit Generalizable Spatial Learners Lu Ling, Yunhao Ge, Yichen Sheng, Aniket Bera CVPR 2026 > proj > arxiv > code Simple and effective formulation with limited data; outperformed SAM3D with unprecedented data scale.
	PixelDiT: Pixel Diffusion Transformers for Image Generation Yongsheng Yu, Wei Xiong†, Weili Nie, Yichen Sheng, Shiqiu Liu, Jiebo Luo CVPR 2026 (Best Paper Finalist) > proj > arxiv An optimal architecture design for pixel space diffusion.
	Scenethesis: Combining Language and Visual Priors for 3D Scene Generation Lu Ling, Chen-Hsuan Lin, Tsung-Yi Lin, Yifan Ding, Yu Zeng, Yichen Sheng, Yunhao Ge, Ming-Yu Liu, Aniket Bera, Max Li ICLR 2026 > proj > arxiv Agentic framework for scaling 3D scene data.
	Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis Yu Yuan, Xijun Wang, Yichen Sheng, Prateek Chennuri, Xingguang Zhang, Stanley Chan CVPR 2025 (highlight) > proj > arxiv A consistency trick for generative computational photography.
	Floating No More: Object-Ground Reconstruction from a Single Image Yunze Man, Yichen Sheng, Jianming Zhang, Liangyan Gui, Yu-Xiong Wang CVPR 2025 > proj > arxiv Pixel height representation improvement.
	Dr.Bokeh: DiffeRentiable Occlusion-aware Bokeh Rendering Yichen Sheng, Zixun Yu, Lu Ling, Zhiwen Cao, Cecilia Zhang, Xin Lu, Ke Xian, Haiting Lin, Bedrich Benes CVPR 2024 > proj > video Fixed the bokeh formula; surpassed Google Pixel Camera with a principled method.
	DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Vision Lu Ling, Yichen Sheng(joint first author)*, Zhi Tu, Wentian Zhao, Cheng Xin, Kun Wan, Lantao Yu, Qianyu Guo, Zixun Yu, Yawen Lu, Xuanmao Li, Xingpeng Sun, Rohan Ashok, Aniruddha Mukherjee, Hao Kang, Xiangrui Kong, Gang Hua, Tianyi Zhang, Bedrich Benes, Aniket Bera CVPR 2024* > proj > arxiv > code A gift for 3D vision community, towards spatial intelligence.
	PixHt-Lab: Pixel Height Based Light Effect Generation for Image Compositing Yichen Sheng, Jianming Zhang, Julien Philip, Yannick Hold-Geoffroy, Xin Sun, HE Zhang, Lu Ling, Bedrich Benes CVPR 2023 (highlight) > proj > arxiv > code Lighting Effects Synthesis Trilogy (SSN, SSG, PixHtLab): mathematical foundation of pixel height representation. Generalize shadow rendering to any lighting rendering with ray-tracing framework in pixel height space.
	Controllable Shadow Generation Using Pixel Height Maps Yichen Sheng* , Yifan Liu* (R.I.P.🕯️), Jianming Zhang, Wei Yin, A. Cengiz Oztireli, He Zhang, Zhe Lin, Eli Shechtman, Bedrich Benes. (Joint first author) ECCV 2022* > proj > arxiv > video > Adobe Max Lighting Effects Synthesis Trilogy (SSN, SSG, PixHtLab): pixel height representation born, shadow rendering in pixel height space. — featured at Adobe MAX 2021
	SSN: Soft Shadow Network for Image Composition Yichen Sheng, Jianming Zhang, Bedrich Benes CVPR 2021 (oral) > proj > arxiv > video > code Lighting Effects Synthesis Trilogy (SSN, SSG, PixHtLab): neural soft shadow rendering.