Xiaobao Wei

I am a second-year dual PhD student at Institute of Software, Chinese Academy of Sciences and Peking University, supervised by Prof. Hui Chen from ISCAS, Prof. Shanghang Zhang from PKU and Ming Lu from Intel Labs China. I received my B.S. in Robotics Engineering from Beihang University in 2023 and obtained Beijing Distinguished Graduate Award.

I serve as a reviewer for prestigious conferences and journals including CVPR, ICLR, NeurIPS, ICME, AISTATS, RA-L, ICSVT.

My areas of focus include neural field, 3D vision.

Email  /  Github  /  Google Scholar

We propose GazeGaussian, a high-fidelity gaze redirection method that uses a two-stream 3DGS model to represent the face and eye regions separately.

We propose a new method called PLGS that enables 3DGS to generate consistent panoptic segmentation masks from noisy 2D segmentation masks while maintaining superior efficiency compared to NeRF-based methods.

We propose a self-supervised street Gaussian (S3Gaussian) method to decompose dynamic and static elements in driving scenes without costly annotations.

We propose I-MedSAM, which leverages the benefits of both continuous representations and SAM, to obtain better cross-domain ability and accurate boundary delineation.

We propose the multi-scale and full spike segmentation network (MFS-Seg), which is based on the deep direct trained SNN and represents the first attempt to train a deep SNN with surrogate gradients for semantic segmentation.

We propose NTO3D, a novel high-quality Neural Target Object 3D (NTO3D) reconstruction method, which leverages the benefits of both neural field and SAM.

We propose DiffusionTalker, a diffusion-based method that utilizes contrastive learning to personalize 3D facial animation and knowledge distillation to accelerate 3D animation generation.

We propose OV-3DET, which leverages advanced image/vision-language pre-trained models to achieve Open-Vocabulary 3D point-cloud DETection.

We propose an end-to-end cross-domain detection Transformer based on the mean teacher framework, MTTrans, which can fully exploit unlabeled target domain data in object detection training and transfer knowledge between domains via pseudo labels.

Object dropping may occur when the robotic arm grasps objects with uneven mass distribution due to additional moments generated by objects gravity. To solve this problem, we present a novel work that does not require extra wrist and tactile sensors and large amounts of experiments for learning.

We investigate group formation-tracking problem for heterogeneous multi-agent systems (HMASs) with both switching networks and communication delays in this paper.