Daoyi Gao

daoyi.gao[at]tum[dot]de

I am a PhD student at the 3D AI Lab at the Technical University of Munich.

Previously, I graduated from the master's program Robotics, Cognition, Intelligence at TUM.

My research interests are in the field of 3D computer vision, CAD-driven scene understanding, and 3D generative modeling.

Publications

arXiv

VoxLution: Evolving Voxels for High-Fidelity 3D Scene Generation

Daoyi Gao*, Xinyi Zhang*, Alexey Bokhovkin, Lei Li, and Angela Dai

arXiv, 2026

Abs Project PDF Video

We introduce VoxLution, a sparse autoregressive cascade that generates high-fidelity 3D indoor scenes by refining a coarse scene structure from 25 cm down to 1.5625 cm voxels.

* Equal contribution.
arXiv

ProcGen3D: Learning Neural Procedural Graph Representations for Image-to-3D Reconstruction

Xinyi Zhang, Daoyi Gao, Naiqi Li, and Angela Dai

arXiv, 2025

Abs Project PDF

We introduce ProcGen3D, a new approach for 3D content creation by generating procedural graph abstractions of 3D objects, which can then be decoded into rich, complex 3D assets.
CVPR25

MeshArt: Generating Articulated Meshes with Structure-Guided Transformers

Daoyi Gao, Yawar Siddiqui, Lei Li, and Angela Dai

CVPR, 2025

Abs Project PDF Code

We propose a hierarchical transformer-based approach to generate articulated 3D meshes with clean, compact geometry.
SIGGRAPH24

DiffCAD: Weakly-Supervised Probabilistic CAD Model Retrieval and Alignment from an RGB Image

Daoyi Gao, Dávid Rozenberszki, Stefan Leutenegger, and Angela Dai

SIGGRAPH, 2024

Abs Project PDF Code

We propose a weakly-supervised approach for CAD model retrieval and alignment from an RGB image. Our approach utilizes diffusion models to tackle the ambiguities in monocular perception, and achieves robust cross-domain performance while being trained only on synthetic data.
ECCV22

Polarimetric Pose Prediction

Daoyi Gao*, Yitong Li*, Patrick Ruhkamp*, Iuliia Skobleva*, Magdalena Wysock*, HyunJun Jung, Pengyuan Wang, Arturo Guridi, and Benjamin Busam

ECCV, 2022

Abs Project PDF Code

We propose a hybrid model that uses polarization information with physical priors in a data-driven learning strategy to improve the accuracy of pose predictions for photometrically challenging objects.

* Equal contribution.
3DV21

Attention meets Geometry: Geometry Guided Spatial-Temporal Attention for Consistent Self-Supervised Monocular Depth Estimation

Daoyi Gao*, Hanzhi Chen*, Patrick Ruhkamp*, Nassir Navab, and Benjamin Busam

3DV, 2021

Abs Project PDF Code

We propose a self-supervised monocular depth estimation pipeline that improves consistency while preserving accuracy, and a Temporal Consistency Metric (TCM) that quantifies depth consistency across frames.

* Equal contribution.

Workshops

Teaching