Yuanze Lin

I am a DPhil student in the Computer Science Department at the University of Oxford, working with Prof. Ronald Clark and Prof. Philip Torr on diffusion and vision-language models.

Before going to Oxford, I had a great time at Microsoft Redmond, MSR Asia, CCVL @ Johns Hopkins University, Alibaba, etc. I appreciate collaborating with distinguished professors and researchers from these institutions.

My research interests span machine learning, especially:

• Multimodal large language models (MLLMs)

• Diffusion-based image/video generation and editing

• Applications of large language models (LLMs)

yuanze.lin [at] cs.ox.ac.uk Google Scholar GitHub LinkedIn

News

[07/2026] Started a research internship at Microsoft Research Cambridge.

[09/2025] IllumiCraft accepted to NeurIPS 2025.

[06/2025] We introduced IllumiCraft for high-fidelity video relighting.

[04/2025] Olympus was selected as a Highlight at CVPR 2025!

[03/2025] Released the code of Olympus (CVPR 2025) for vision tasks.

[02/2025] Olympus accepted to CVPR 2025.

[12/2024] We presented Olympus to solve over 20 different computer vision tasks.

[08/2024] Released the code of Learnable Regions (CVPR2024) for image editing.

[07/2024] Rethinking Visual Prompting for MLLMs with External Knowledge was presented.

[03/2024] Check out DreamPolisher for high-quality text-to-3D generation!

[02/2024] Started a research internship at GenAI @ Microsoft.

[02/2024] Text-Driven Image Editing via Learnable Regions accepted to CVPR 2024.

[10/2023] Started my DPhil journey at CS @ University of Oxford.

[07/2023] SMAUG accepted to ICCV 2023.

[09/2022] REVIVE accepted to NeurIPS 2022.

[03/2022] Pseudo-Q and AdaFocus V2 accepted to CVPR 2022.

[07/2021] MCN accepted to ICCV 2021.

[06/2021] EVA-GCN accepted to CVPR 2021 AMFG Workshop and won 🏆 Best Paper Award

Selected Publications

Papers are sorted by recency, * denotes equal contribution.

	IllumiCraft: Unified Geometry and Illumination Diffusion for Controllable Video Generation Yuanze Lin, Yi-Wen Chen, Yi-Hsuan Tsai, Ronald Clark, Ming-Hsuan Yang NeurIPS, 2025 ArXiv / Project Page / Video / Code / BibTeX We present IllumiCraft, a unified framework that unifies geometry and illumination diffusion for controllable video generation.
	Olympus: A Universal Task Router for Computer Vision Tasks Yuanze Lin, Yunsheng Li, Dongdong Chen, Weijian Xu, Ronald Clark, Philip Torr CVPR, 2025 ★ Highlight ArXiv / Project Page / Video / Poster / Code / BibTeX Turns MLLMs into a universal task router that handles a wide array of computer vision tasks within a single unified framework.
	Text-Driven Image Editing via Learnable Regions Yuanze Lin, Yi-Wen Chen, Yi-Hsuan Tsai, Lu Jiang, Ming-Hsuan Yang CVPR, 2024 ArXiv / Project Page / Video / Poster / Code / BibTeX A region-based network trained with a CLIP-guided text-driven loss, editing images from freely provided language descriptions.
	SMAUG: Sparse Masked Autoencoder for Efficient Video-Language Pre-training Yuanze Lin, Chen Wei, Huiyu Wang, Alan Yuille, Cihang Xie ICCV, 2023 ArXiv / Poster / Slides / BibTeX An efficient video-language pre-training framework that stays competitive on retrieval and video QA while cutting pre-training cost by 1.9X or more.
	REVIVE: Regional Visual Representation Matters in Knowledge-Based Visual Question Answering Yuanze Lin, Yujia Xie, Dongdong Chen, Yichong Xu, Chenguang Zhu, Lu Yuan NeurIPS, 2022 ArXiv / Poster / Supplementary Material / OpenReview / Code / BibTeX A knowledge-based VQA method exploiting explicit object-region information in both retrieval and answering, reaching state-of-the-art on OK-VQA.
	Pseudo-Q: Generating Pseudo Language Queries for Visual Grounding Haojun Jiang, Yuanze Lin, Dongchen Han, Shiji Song, Gao Huang CVPR, 2022 ArXiv / Poster / Code / BibTeX Automatically generates pseudo language queries for supervised training, matching or beating weakly-supervised visual grounding across five datasets.
	AdaFocus V2: End-to-End Training of Spatial Dynamic Networks for Video Recognition Yulin Wang, Yang Yue, Yuanze Lin, Haojun Jiang, Zihang Lai, Victor Kulikov, Nikita Orlov, Humphrey Shi, Gao Huang CVPR, 2022 ArXiv / Code / BibTeX Reformulates AdaFocus as a one-stage algorithm via differentiable patch selection and an improved training scheme, validated on six benchmarks.
	Self-supervised video representation learning with meta-contrastive network Yuanze Lin, Xun Guo, Yan Lu ICCV, 2021 ArXiv / Poster / BibTeX A Meta-Contrastive Network combining contrastive and meta-learning for pre-training, surpassing prior methods on UCF101 and HMDB51.
	EVA-GCN: Head Pose Estimation Based on Graph Convolutional Networks Miao Xin, Shentong Mo, Yuanze Lin CVPR AMFG Workshop, 2021 🏆 Best Paper Award Paper / Code / BibTeX Builds a landmark-connection graph and uses Graph Convolutional Networks to model nonlinear mappings from graph topology to head-pose angles.