Daily arXivWei Deng 邓韦
Stay hungry, stay foolish, stay lazy!
Wei Deng is currently a third-year Ph.D. student at MIC-Lab, BUPT, advised by Prof. Mengshi Qi.
His research interests include:
- Computer vision
- Embodied reasoning, navigation, manipulation and scene generation
He is also passionate about computer science and technology. He welcomes discussing research and technology topics with him: dw-dengwei@bupt.edu.cn.
News
[2026.5.14] Honor to receive the ICML 2026 Silver Reviewer Award.
[2026.5.01] One paper is accepted by ICML 2026.
[2025.2.27] One paper is accepted by CVPR 2025.
Publications
* indicates the corresponding author(s), indicates equal contributions
Mengshi Qi*, Wei Deng, Xianlin Zhang, Huadong Ma. Global-Local Monte Carlo Tree Search in Vision-Language Models for Text-to-3D Indoor Scene Generation. arXiv 2026 (Extension of our CVPR 2025). [PDF] [Code]
In this paper, we propose a novel Global-Local Monte Carlo Tree Search in Vision-Language Models for Text-to-3D Indoor Scene Generation.
Wei Deng, Xianlin Zhang, Mengshi Qi*. Active Exploring like a Pigeon: Reinforcing Spatial Reasoning via Agentic Vision-Language Models. ICML 2026. [PDF] [Code]
Inspired by pigeons’ building and exploiting cognitive maps for navigation, in this work we propose a novel agentic pipeline for spatial reasoning with VLMs.
Wei Deng, Mengshi Qi*, Huadong Ma. Global-Local Tree Search in VLMs for 3D Indoor Scene Generation. CVPR 2025. [PDF] [Code]
This paper considers 3D indoor scene generation as a planning problem subject to spatial and layout common sense constraints. To solve the problem with a VLM, we propose a new global-local tree search algorithm.
Lixiong Qin, Mei Wang, Xuannan Liu, Yuhang Zhang, Wei Deng, Xiaoshuai Song, Weiran Xu*, Weihong Deng, Faceptor: A generalist model for face perception. ECCV 2024 Oral. [PDF] [Code]
Faceptor is a unified face perception generalist that uses task queries to achieve competitive or superior performance to task-specific models across diverse face analysis tasks.
Services
Reviewing
- Conference: ICML, NeurIPS, AAAI, ICME