Xiao-Ming Wu (伍晓鸣)

M.S. student

School of Computer Science and Engineering

Sun Yat-sen University

wuxm65@mail2.sysu.edu.cn

Scholar  /  Github

Biography

I'm currently a second-year master student at Sun Yat-sen University, advised by Prof. Wei-Shi Zheng, where I cultivate the interest in research, and develop the scientific ability and taste of it. Previously, I obtain my B.E. degree in Shandong University. At that time, I had my first attempt on scientific research, advised by Xin Luo and Xin-Shun Xu. I am now looking for a potential Ph.D. position in 2025 Fall.

Research Interests

Research is for curiosity and fun.

Now I am interested in how to build a robotics system that unifies learning and control, which aims to fuse representation learning into each part in robotic modeling and control, making robotics control able to perceive and understand. Thus, my current research interests include:

• Robotics and Embodied AI
• Deep Representation Learning

Besides, I have been working on many other topics, including diffusion models, low-level vision and hashing retrieval. Participating in diverse research areas broadens my horizon about deep learning.

Publications

Below are my publications (show by date). My first author works are highlighted.

Paper and code will be ready soon.

A new transformer-based framework for dexterous grasp generation, capable of predicting a diverse set of feasible grasp poses only in one pass.

Exploring a new task of generating human grasps based on single-view scene point clouds. A new baseline and a new dataset are also proposed for this novel task.

A diffusion-based universal image restoration model, with an assemble-then-separate (like the hourglass) mapping for multi-tasks training.

A diffusion-based module to gradually remove the redundancy in the cost volume for stereo matching, which is also plug-and-play for volume-based methods.

A new perspective to view the binary neural network training: equilibrium between estimating error and gradient stability. And a simple and effective gradient estimator is proposed to balance it.

A prompt-based feature mapping to generate more real-domain anomalies to supervise the anomaly detection training.

Generating pseudo tags based on the co-occurrence similarity between tags to handle the weakly-supervised online hashing.

Introducing semantic into online hashing, handling the dimension mismatching and mitigating the inconsistent problem in incremental online hashing.

Inspired by Discrete Latent Factor Hashing (DLFH), we propose an online-version DLFH, with many novel designs to handle the online problem in cross-modal hashing.

Academic Service

Journal Reviewer: Pattern Analysis and Machine Intelligence (TPAMI), Pattern Recognition (PR).

Conference Reviewer: Computer Vision and Pattern Recognition (CVPR) 2024, ACM Multimedia (MM) 2024.

Awards