Stefan Lionar

I am a PhD student in Computer Science at the National University of Singapore (NUS), supervised by Prof. Gim Hee Lee and under the Industrial PhD Programme with Garena and Sea AI Lab. My work focuses on creating high-quality 3D digital worlds and embodied agents that interact within them. I obtained my Master’s degree in Robotics from ETH Zurich, and Bachelor’s degree in Mechanical Engineering from Nanyang Technological University.

Email | GitHub | Google Scholar | OpenReview | LinkedIn

News

[02.2026] TeamHOI is accepted to CVPR 2026! One paper on mesh generation is also recommended for CVPR Findings track!
[10.2025] Selected as a top reviewer (top 8%) at NeurIPS 2025!
[08.2025] Received my second Research Achievement Award from NUS School of Computing!
[02.2025] TreeMeshGPT is accepted to CVPR 2025!
[01.2024] Received Research Achievement Award from NUS School of Computing!
[09.2023] NU-MCC is accepted to NeurIPS 2023!
[01.2023] Started my PhD study at NUS.
[10.2021] My master's thesis (NeuralBlox) is accepted to 3DV 2021!
[11.2020] Our 3D vision course project is accepted to WACV 2021!

Publications

	TeamHOI: Learning a Unified Policy for Cooperative Human-Object Interactions with Any Team Size Stefan Lionar, Gim Hee Lee Computer Vision and Pattern Recognition (CVPR), 2026 Paper / Project Page / Code A novel framework for scalable cooperative human-object interaction.
	TreeMeshGPT: Artistic Mesh Generation with Autoregressive Tree Sequencing Stefan Lionar, Jiabin Liang, Gim Hee Lee Computer Vision and Pattern Recognition (CVPR), 2025 Paper / Code / Demo / Video An autoregressive artistic mesh generation method that retrieves the next token from a dynamically growing tree structure. It achieves a state-of-the-art compression rate.
	NU-MCC: Multiview Compressive Coding with Neighborhood Decoder and Repulsive UDF Stefan Lionar, Xiangyu Xu^✉, Min Lin, Gim Hee Lee Neural Information Processing Systems (NeurIPS), 2023 (^✉ Corresponding author) Paper / Project Page / Code / Video An efficient architecture and representation for single-view 3D reconstruction.
	NeuralBlox: Real-Time Neural Representation Fusion for Robust Volumetric Mapping Stefan Lionar, Lukas Schmid, Cesar Cadena, Roland Siegwart, Andrei Cramariuc International Conference on 3D Vision (3DV), 2021 (*Equal contribution) Paper / Code A robust incremental volumetric mapping framework using neural representation.
	Dynamic Plane Convolutional Occupancy Networks Stefan Lionar, Daniil Emtsev, Dusan Svilarkovic, Songyou Peng Winter Conference on Applications of Computer Vision (WACV), 2021 (Equal contribution) Paper / Code / Video We propose implicit representation using multiple 2D planes for accurate 3D surface reconstruction from point cloud.

Other Selected Projects

	2.5D U-Net for Fast and Accurate Brain Tumor Segmentation BioMind 2022
	Robust Object Detection in Duckietown Maximilian Stölzle, Stefan Lionar Duckietown class at ETH Zurich, 2019 Explore various data augmentation, design safety-oriented loss function, and integrate Google Coral Edge TPU for real-time inference.

Website template is taken from here