Senior Researcher in 3D vision, language, and embodied AI
Guofeng (Ryan) Mei
Senior Researcher at Technologies of Vision (TeV), Fondazione Bruno Kessler (FBK).
I develop methods that connect geometry, semantics, and real-world interaction, spanning 3D vision-language models, world models, and embodied AI.
"Never underestimate your power to change yourself!"
Research Overview
How can AI understand a 3D world, describe it in language, and use that understanding to act? My research connects geometric learning with multimodal perception, including learning representations without manual annotations and applying them to robotic interaction.
3D Vision & Language
I build models that connect point clouds with language for scene understanding, question answering, and semantic grounding. A central goal is to make 3D multimodal learning efficient and adaptable.
Point Cloud Registration
I develop methods to align partially overlapping 3D point clouds. UDPReg combines probabilistic representations and distribution-level matching to learn registration without ground-truth transformations.
Unsupervised Learning
I develop methods that learn transferable point cloud representations without manual annotations. CluRender combines soft clustering with neural rendering to learn from 3D geometry and 2D appearance.
Embodied Perception & Interaction
I explore how 3D perception supports robotic reasoning and grasping in complex scenes. My work connects scene understanding with decisions about how to interact with and retrieve objects.
3D Generation & World Models
I study generative models that capture 3D scene structure and support controllable synthesis. T2LDM uses diffusion with self-conditioned representation guidance to generate detailed LiDAR scenes from text.
I earned my Ph.D. at the University of Technology Sydney in 2023, supervised by Prof. Jian Zhang, and now work with Dr. Fabio Poiesi at FBK. More about my background
Selected Work
All publicationsOne representative paper for each research theme, with an emphasis on my first-author and co-first-author work.
† Equal contribution · * Corresponding author
3D Vision & Language
Efficient Encoder-Free Fourier-based 3D Large Multimodal Model
Fourier-based point cloud tokens connect 3D geometry to a language model without a separate 3D encoder.
First author
Embodied Perception & Interaction
Obstruction reasoning for robotic grasping
UNOGrasp reasons about objects that obstruct a target and predicts grasps to support object retrieval in cluttered scenes.
Co-author
3D Generation & World Models
A Self-Conditioned Representation Guided Diffusion Model for Realistic Text-to-LiDAR Scene Generation
Self-conditioned representation guidance helps a text-to-LiDAR diffusion model generate realistic 3D scenes with detailed objects and controllable scene content.
Second author
Unsupervised Learning
Unsupervised Point Cloud Representation Learning by Clustering and Neural Rendering
Soft clustering and neural rendering learn transferable point cloud features from geometry and image appearance, without manual annotations or data augmentation.
First author
Point Cloud Registration
Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration
Gaussian mixture representations and distribution-level correspondences align partially overlapping point clouds without ground-truth transformation supervision.
First author
News
| 2026-10 | Recognized as a NeurIPS 2026 Top Reviewer. |
|---|---|
| 2026-10 | Latent Riemannian Flow Matching for Geometry-Grounded 3D Foundation Models was accepted to NeurIPS 2026. |
| 2026-08 | Gave an invited talk, “Can AI Understand and Predict the Physical World?”, at Shandong University Summer School. |
| 2026 | New CVPR 2026 work includes Fase3D, GLASS, and UNOGrasp. |
| 2026 | Received ECCV Outstanding Reviewer and ICML Silver Reviewer recognition. |
| 2025-06 | Papers accepted to ICCV 2025 and IROS 2025. |
| 2025-02 | PerLA was accepted to CVPR 2025. |
| 2024-11 | Two papers were accepted to 3DV 2025. Congratulations to all my collaborators! |
| 2024-09 | One paper was accepted to ACCV 2024. Congratulations to all my collaborators! |
| 2024-07 | One paper was accepted to TCSS 2024. Congratulations to all my collaborators! |
| 2024-07 | One paper was accepted to ICPR 2024. Congratulations to all my collaborators! |
| 2024-02 | Two papers were accepted to CVPR 2024. Congratulations to all my collaborators! |
| 2024-02 | One paper was accepted to IJCV 2024. Congratulations to all my collaborators! |
| 2024-01 | One paper was accepted to ICRA 2024. Congratulations to all my collaborators! |
| 2023-03 | One paper was accepted to CVPR 2023. Congratulations to all my collaborators! |
| 2022-10 | One paper was accepted to BMVC 2022 (Oral). Congratulations to all my collaborators! |
| 2022-02 | One paper was accepted to ICME 2022 (Oral). Congratulations to all my collaborators! |
Service & Recognition
All service activitiesReviewer Recognition
- Top ReviewerNeurIPS · 2024, 2026
- Outstanding ReviewerECCV · 2026 / ICCV · 2025 / CVPR · 2024, 2025
- Silver ReviewerICML · 2026
Academic Community
I review for conferences including CVPR, ICCV, ECCV, NeurIPS, ICML, and ICLR, and journals including IEEE TPAMI, TIP, and Pattern Recognition.
Organizer of the NTIRE 2026 Efficient Super-Resolution Challenge at CVPR Workshops.
Invited talks