Xuexun Liu

I am a third year master student in the College of Computer Science and Software Engineering at Shenzhen University, advised by Prof. Xu Wang . In 2023, I obtained my B.Eng. in the College of MechaTronics and Control Engineering, Shenzhen University.

My research interests lie in computer vision and multimodal learning. Currently, I am focusing on label-efficient and multimodal 3D scene understanding and fine-tuning vision-language models (VLMs) for specialized tasks.

From May to September 2025, I worked as a Computer Vision Algorithm Intern at Ant Group.

From December 2025 to March 2026, I worked as a Deep Learning Algorithm Intern at Insta360.

Email / Google Scholar / Github

News

2024-09: Our work, LESS, on referring 3D segmentation has been accepted at NeurIPS 2024.

2025-11: Our work, DBGroup, on weakly supervised 3D segmentation has been accepted at AAAI 2026 (Oral).

2026-01: Our work, Interactive360°, on interactive object removal in 360° panoramic images has been accepted to TMM 2026.

Publications

* indicates equal contribution

	LESS: Label-Efficient and Single-Stage Referring 3D Segmentation Xuexun Liu, Xiaoxu Xu, Jinlong Li, Qiudan Zhang , Xu Wang , Nicu Sebe , Lin Ma Conference on Neural Information Processing Systems (NeurIPS)*, 2024 [arXiv] [Code] [Project Page]
	DBGroup: Dual-Branch Point Grouping for Weakly Supervised 3D Semantic Instance Segmentation Xuexun Liu, Xiaoxu Xu, Qiudan Zhang , Lin Ma , Xu Wang Association for the Advancement of Artificial Intelligence (AAAI(Oral)), 2026 [arXiv] [Code]

Projects

	Model Compression and Efficient Reasoning Xuexun Liu et al. Tencent Kaiwu Artificial Intelligence National Competition (5th place nationwide), 2023.11 [Competition Link] Implemented model compression techniques on neural networks incorporating CNN, LSTM, and Transformer architectures. Designed a lightweight student model, implementing knowledge distillation. Applied pruning, quantization, and fine-tuning methodologies, resulting in a model compressed to 10% of its original size while maintaining equivalent performance metrics.
	Unofficial implementation of "RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds" Xuexun Liu Open-source Project, 2023.05 [Code] An unofficial pytorch implementation of "RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds"
	Automatic Balance Small Bike Design Xuexun Liu et al. National College Student Intelligent Car Competition, National Finals, the First Prize, 2022.08 [Competition Link] Implemented I2C protocol to acquire gyroscope accelerometer data and applied Kalman filtering for attitude calculation. Designed cascade PID controllers for balancing and driving wheel DC motors. Developed a vision system using grayscale camera inputs with binarization and morphological processing. Applied Sobel edge detection to identify track features and generate optimal trajectories.
	Robot Navigation Simulation Xuexun Liu Curriculum Design & Open-source Project, 2022.06 [Code] Implemented robot navigation utilizing artificial potential field algorithm and its enhanced version. Designed and executed comprehensive simulations in MATLAB.
	BoxMan in JZ2440 Xuexun Liu Curriculum Design & Open-source Project, 2021.12 [Code] Developed LCD display interface programming, implemented ADC functionality to process touch screen inputs, and engineered image collision detection algorithms with corresponding collision response logic.

Honors and Awards

2023 Outstanding Graduates, Shenzhen University

2022 National Scholarship

Website Template