PhD in Computer Science

The University of Adelaide


Hao Chen is currently a CV researcher @ Noah’s Ark Lab in Huawei. He got his PhD degree of computer science in Adelaide Intelligent Machines Group at the University of Adelaide. He was advised by Professor Chunhua Shen. Before Adelaide he obtained his undergraduate and master degree from Zhejiang University, and had been a researcher in NetEase Inc.

He anticipates to graduate in late 2020.

His general research interest lies in probabilistic methods for general perception tasks in computer vision. His recent research involves:

  • Fast NAS: computation-friendly neural architecture search (NAS) with reinforcement learning for segmentation, detection and low-level tasks.
  • BlendMask: a general instance segmentation framework which is both faster and more efficient than Mask-RCNN.
  • AdelaiDet: a collection of recent projects dealing with different instance-level tasks.


  • Machine Learning
  • Computer Vision
  • Bayesian Deep Learning


  • PhD in Artificial Intelligence, 2020

    The University of Adelaide

  • MSc in Computer Science, 2016

    Zhejiang University

  • BSc in Computer Science, 2013

    Zhejiang University

Recent Posts

Four Papers Got Accepted at CVPR 2020

All papers are available on arxiv.

Faster and Finer Instance Segmentation With Blendmask

Introduction to our new instance segmentation model, BlendMask, which generates higher qualities than Mask R-CNN while being faster in inference.

NAS - Where Are We Now

The reasons why people are either running towards or away from Neural Architecture Search.

On Optimization in Deep Learning

An old post about no local minima in Deep Learning.



AdelaiDet is AIM’s research platform for instance-level detection tasks based on Detectron2. A collection of recent work from the AIM group based on modern infrastructure.


Code for ABCNet: Real-time Scene Text Spotting with Adaptive Bezier-Curve Network, CVPR ‘20. End-to-end text recognition from the wild with Bezier points.


Code for NAS-FCOS: Fast Neural Architecture Search for Object Detection, CVPR ‘20. Computation-friendly NAS for object detection.


Code for Fast Neural Architecture Search of Compact Semantic Segmentation Models via Auxiliary Cells, CVPR ‘19. RL based NAS for semantic segmentation with bags of tricks.

Recent & Upcoming Talks


An invited talk in Chinese about CVPR20 publication BlendMask.

Recent Publications

Quickly discover relevant content by filtering publications.

Deep Learning for Source Code Modeling and Generation: Models, Applications and Challenges

I wrote the paper then Triet helped me with the revisions. A survey of deep language models and their applications in software engineering. Paper got accepted by Computing Survey.

BlendMask: Top-Down Meets Bottom-Up for Instance Segmentation

Contributed equally with the second author. In this work, we achieve improved mask prediction by effectively combining instance-level information with semantic information with lower-level fine-granularity. Our main contribution is a blender module which draws inspiration from both top-down and bottom-up instance segmentation approaches. The proposed BlendMask can effectively predict dense per-pixel position-sensitive instance features with very few channels, and learn attention maps for each instance with merely one convolution layer, thus being fast in inference. BlendMask can be easily incorporated with the state-of-the-art one-stage detection frameworks and outperforms Mask R-CNN under the same training schedule while being 20% faster.

Memory-Efficient Hierarchical Neural Architecture Search for Image Denoising

Haokui adapted my implementation of AutoDeepLab for low-level image processing tasks. One highlight of the framework is that it save two thirds of the memory required for AutoDeepLab.

NAS-FCOS: Fast neural architecture search for object detection

Contributed equally with the first two authors. We built upon our CVPR19 work for object detection on COCO. At first I thought it would be easy but we met many obstacles mostly about training efficiency. Even up to now I cannot say that we have dealt with all of those. However, this is what we got after grinding for almost a year.

Architecture Search of Dynamic Cells for Semantic Video Segmentation

Contributed equally with the first author. In this work we attempt to come up with generalisation of dynamic cells in video segmentation, and instead of manually designing contextual blocks that connect per-frame outputs, we propose a neural architecture search solution, where the choice of operations together with their sequential arrangement are being predicted by a separate neural network.


  • The Australian Institute for Machine Learning, Adelaide, SA 5000
  • Monday 10:00 to 13:00
    Wednesday 09:00 to 10:00
  • Book an appointment