Wanli Peng

I received the B.E. degree (ranking first in major) in electronic and information engineering from the Dalian University of Technology (DLUT) , Dalian, China in 2018. I am currently a PhD student in signal and information processing from DLUT. My advisor is Prof. Yi Sun.

My research interests include 2D & 3D object detection, 6Dof pose estimation, Shape reconstruction and Robotic manipulation.

Email: 1136558142@mail.dlut.edu.cn


TransGrasp: Grasp Pose Estimation of a Category of Objects by Transferring Grasps from Only One Labeled Instance

We propose TransGrasp, a category-level grasp pose estimation method that predicts grasp poses of a category of objects by labeling only one object instance.

Hongtao Wen, Jianhang Yan, Wanli Peng*, Yi Sun
ECCV, 2022
Webpage  •   Paper  •   Code

Self-Supervised Category-Level 6D Object Pose Estimation with Deep Implicit Shape Representation

A self-supervised method for category-level 6D pose estimation, SSC-6D, which can predict unseen object poses without explicit pose annotations and exact 3D models in real scenarios for training.

Wanli Peng, Jianhang Yan, Hongtao Wen, Yi Sun*
AAAI, 2022
Webpage  •   Paper  •   Code

IDA-3D: Instance-Depth-Aware 3D Object Detection from Stereo Vision for Autonomous Driving

An end-to-end learning framework for 3D object detection based on stereo images in autonomous driving.

Wanli Peng, Hao Pan, He Liu, Yi Sun*
CVPR, 2020
Webpage  •   Paper  •   Code


Demonstrations for robot manipulation based on TransGrasp

We built a complete robot manipulation pipeline based on ROS, where where we use our TransGrasp to predict robust grasp poses for robotic manipulation.

  • Robot-assisted watering mobility-impaired individuals.
  • Autonomously pouring water from cup into bowl.
  • Autonomously grasping household objects.


A Defect Inspection System for Critical Component of Automobile
  • An automatic defect detection software based on X-ray real-time imaging.
  • Automatic unattended operation of image acquisition, defect detection and workpiece sorting.
  • Enhance details of weak targets using the MUSICA algorithm.
  • Reimplement the forward propagation of Deep Learning algorithm based on CUDA.
  • Achieve less than 1% false negative rate in the actual production environment.