Kangle Deng
Kangle Deng is a computer vision researcher who builds generative models for 3D content. After a B.S. at Peking University in 2020, he completed a PhD in Carnegie Mellon's Robotics Institute in August 2025, co-advised by Jun-Yan Zhu and Deva Ramanan. During the doctorate he won a 2022 Microsoft Research PhD Fellowship and coauthored Depth-supervised NeRF in 2022, pix2pix3D in 2023 and FlashTex in 2024. BrickGPT, a system that generates physically stable brick structures from text, won the Marr Prize at ICCV 2025. He is a research scientist at Roblox, where his work focuses on generative 3D content creation and world models.
Timeline
-
2020
A Peking University degree
He earns a B.S. from Peking University in 2020, before joining Carnegie Mellon.
-
2022
DS-NeRF and a Microsoft fellowship
Depth-supervised NeRF, a CVPR 2022 paper written with Andrew Liu, Jun-Yan Zhu and Deva Ramanan, proposes DS-NeRF, which learns neural radiance fields with depth supervised by 3D point clouds. In 2022, Microsoft Research also selects him as one of 36 students worldwide for its PhD Fellowships; he is a Robotics Institute PhD student working on computer vision and graphics.
-
2023
The pix2pix3D model
The CVPR 2023 paper 3D-aware Conditional Image Synthesis, coauthored with Gengshan Yang, Deva Ramanan and Jun-Yan Zhu, introduces pix2pix3D. The model takes a 2D label map, such as a segmentation or edge map, and synthesizes a corresponding image from different viewpoints.
-
2024
FlashTex mesh texturing
FlashTex, an ECCV 2024 paper coauthored with Timothy Omernick, Alexander Weiss, Deva Ramanan, Jun-Yan Zhu, Tinghui Zhou and Maneesh Agrawala, textures an input 3D mesh from a text prompt. The generated textures relight properly in different lighting environments.
-
March 2025
Roblox Cube foundation model
In March 2025, Roblox's Foundation AI Team posts Cube: A Roblox View of 3D Intelligence, with him among the authors. The paper presents a first step toward a foundation model for 3D intelligence, built on a 3D shape tokenizer that supports text-to-shape, shape-to-text and text-to-scene generation.
-
August 2025
A Carnegie Mellon PhD
In August 2025, he finishes his PhD in the Robotics Institute at Carnegie Mellon, co-advised by Jun-Yan Zhu and Deva Ramanan. The thesis is titled Learning to Create 3D Content: Geometry, Appearance, and Physics.
-
2025
BrickGPT wins the Marr Prize
BrickGPT, which he coauthors with Ava Pun, Ruixuan Liu, Deva Ramanan, Changliu Liu and Jun-Yan Zhu, wins the Best Paper Award (Marr Prize) at ICCV 2025. It is the first approach for generating physically stable interconnecting brick assembly models from text prompts, using an autoregressive language model that predicts the next brick to add. A demo produces step-by-step build guides for 21 types of models, including a birdhouse, sofa and piano.
-
As of October 2026
Research scientist at Roblox
He is a research scientist at Roblox, where his research focuses on generative 3D content creation and world models. He is an author of CubePart, a SIGGRAPH 2026 paper on an open-vocabulary, part-controllable 3D generator.
Career
- NowRobloxResearch scientist
Education
- 2025Carnegie Mellon UniversityPhD
- 2020Peking UniversityB.S.
Sources and credits
Photo: We have no published photo of Kangle Deng yet. Send us a free-licensed photo.
- Kangle Deng
- SCS Students Named Microsoft Research Ph.D. Fellows
- Microsoft Research PhD Fellowship - Microsoft Research
- [2503.15475] Cube: A Roblox View of 3D Intelligence
- Kangle Deng - Robotics Institute Carnegie Mellon University
- ICCV 2025 Open Access Repository
- BrickGPT won the Best Paper Award (Marr Prize) at ICCV 2025
- The AI Tool That Could Make Manufacturing Faster and More Efficient — by Using Lego Bricks - News - Carnegie Mellon University
- CubePart: An Open-Vocabulary Part-Controllable 3D Generator | Roblox

