Dr Hang Dai

  • Lecturer in Machine Learning (School of Computing Science)

Biography

News

  • 19/11/2022 One paper is accepted in AAAI 2023. Congrats to former Intern Xiaobin Hu.
  • 08/07/2022 Three papers are accepted in ECCV 2022. Congrats to former Intern Jiale Li, Yu Hong, and Postdoc Dr. Xuebin Qin. Papers: CMKD (Oral presentation, acceptance rate: 2.7%), Self-Distillation,  DIS.

Bio

Dr. Hang Dai is a Lecturer in Machine Learning at the School of Computing Science (IDA Section), University of Glasgow, Scotland, UK. He holds a Ph.D. in Computer Science from the University of York, where he built a 3D morphable model of human heads (LYHM) from the Headspace dataset. He was awarded a three-year full Oversea Research Scholarship from the University of York during his Ph.D. study.

He is a regular publisher in top AI/CV conferences like CVPR, ECCV, ICCV, ACM MM and etc. He also wins competitive top conference challenge awards, such as CVPR 2022 Waymo Autonomous Driving Challenge and ECCV 2020 Command for Autonomous Vehicles Challenge. He leads his research group to formulate the problem of highly accurate dichotomous image segmentation towards entering an era of demanding highly accurate outputs from AI algorithms to support delicate human-machine interaction and immersed virtual life and deliver the DIS benchmark to the community.

                           Highly Accurate Dichotomous Image Segmentation

Research interests

His research interests lie in the intersection of computer vision, pattern recognition, and artificial intelligence.

Research Interests.

3D Computer Vision: 3D shape analysis, 3D from X, face manipulation from 3D.

Autonomous Driving: 3D semantic segmentation, 3D object detection, Occupancy network, Interactive Planning, Lanes Network, Autolabeling, Simulation.

Low-level Vision Task: Object referral, Highly accurate image segmentation in Natural images, and Bio-medical images.

Ph.D. Vacancies.

I am looking for highly motivated and competitive Ph.D. candidates. Publications in top conferences like CVPR, ECCV, and ICCV are a PLUS. For Ph.D. topics, please refer to the section of Supervision. If you are interested, drop me an email with your CV. I will reply when the Ph.D. applicant meets the requirements. 

[Hot!] One full Ph.D. studentship at Home Rate is available for the coming academic year (23-24). Qualified international students need to pay for the fee differences. This position is very competitive. Application Deadline: January 2023.

[CSC Ph.D. Studentship] Deadline: January of the academic year.

• What is covered: if accepted by the CSC, the CSC covers the stipend. The college’s graduate school endeavors to match this with a tuition waiver.
• Eligibility of Student: student has to be Chinese and has to agree to the CSC’s terms which include the requirement to return to China after completing their Ph.D.

Students who are not already in Glasgow, and who are affiliated with partner institutions are encouraged to apply for CSC Ph.D. studentship.

[Other scholarships] Deadline is usually January of the academic year. CDT studentships’ deadline is decided by the CDT team. More scholarship opportunities can be found here: University of Glasgow - Schools - School of Computing Science - Postgraduate research - Prospective students - PhD projects and funding opportunities.

[Ph.D. application process] Please refer to University of Glasgow - Schools - School of Computing Science - Postgraduate research - Prospective students. For international students, please make sure that you have the documentation that you meet the language requirements for the College of Science & Engineering before you start the application process.

 

Supervision

Ph.D. projects are not limited to the following aspects:

3D Computer Vision:

  • Semantic 3D face reconstructions from a single image
  • 3D hand reconstruction/tracking from RGB image or combined sensor data
  • 3D face recognition from combined sensor data
  • Deepfake generation or detection using 3D face reconstruction

 

                                    3D face reconstructions from a single image         

Autonomous Driving:

  • 3D semantic segmentation from point cloud (Self-Distillation) or sensor fusion (LiDAR+RGB).
  • 3D object detection from monocular image (M3dssd, Pseudo-Stereo, CMKD), stereo image (DSGN), point cloud (voxel-to-point, Anchor-free 3DSSD), or sensor fusion (LiDAR+RGB).
  • Occupancy network (MonoScene)
  • Interactive Planning: Optimization-based Trajectory Planner with constraints generated incrementally over search steps.
  • Lanes Network (HDMapGen, VectorMapNet)
  • Auto-labeling the captured sensor data using semi-supervised learning or weakly-supervised learning.  
  • Simulation: Sim2Real for autonomous driving.

                                        Monocular 3D Object Detection

Low-level vision task:

  • Object referral for Salient Object Detection, Commands for Autonomous Vehicles (1st, 2nd place).
  • Highly accurate image segmentation in natural  (including semantic segmentation, instance segmentation, salient object detection, and camouflaged object detection), and bio-medical images (for example, retina vessel segmentation).

 

Teaching

COMPSCI5012 Internet Technology M - 2022-23