Congratulations! AIML PhD student awarded prestigious 2021 Google PhD Fellowship

Xinlong Wang

Xinlong Wang

Story written by Dr Sarah Keenihan, AIML

University of Adelaide and Australian Institute for Machine Learning (AIML) PhD student Xinlong Wang has been awarded the 2021 Google PhD Fellowship in Machine Perception, Speech Technology and Computer Vision. 

This year, Wang is one of only four students across Australia to be recognised in the prestigious Google awards for early-stage postgraduate researchers. 

He will receive a monetary prize of A$15,000 to cover costs of research-related activities, and a Google Mentor who will provide guidance on technological capability and career development. 

Under the supervision of Professor Chunhua Shen at AIML, Wang’s PhD research focuses on a challenge in computer vision known as instance segmentation. 

He’s developing a new method for instance segmentation that could save computer vision researchers and technology companies considerable time and money in the future. 

Distinguishing one object from another

Object recognition and localisation is vital for robots to be able to identify objects when they’re moving through a building, and for self-driving cars to avoid pedestrians on city streets. A key part of this capability requires instance segmentation. 

“Instance segmentation is a fundamental problem in computer vision,” Wang explains. 

“It aims to localise each object in an image and place it in a category – for example, these pixels over here all belong to a tree, and those pixels over there all are part of a person.” 

Currently, instance segmentation is a significant time and money investment, as the computer model does not innately know the difference between a tree and person – it has to be taught. 

“When we train models like this, we have to use images that are annotated, or labelled,” Wang says. 

It’s the process of labelling images that adds the time and cost.  

“And so my work aims to develop a new approach for instance segmentation that has comparable accuracy, but only requires less than one third of the annotation of existing methods,” Wang explains. 

An example of a computer vision training system might consist of 1 million images – and an existing instance segmentation approach would require annotation of every single one of these. 

By contrast, Wang’s approach would still work with the 1 million images, but only 20-30% would need to be labelled to achieve the same accuracy of training. 

A bright future ahead

Having started his postgraduate studies in 2020, Wang hopes to complete his PhD by 2023. 

“AIML provides a great environment for young researchers like me,” he says. 

“I’m very grateful for this award, and to my supervisor Professor Shen for providing an opportunity to work with him.” 

Wang has also taken place in several industry placements as part of his career development. 

He hopes to continue working in computer vision into the future, perhaps as a researcher in the private sector or in a university. 

Tagged in Award, AIML, Computer Vision