Guocheng Qian

钱国成

Ph.D. student in Computer Science

King Abdullah University of Science and Technology (KAUST)

Guocheng Qian

Guocheng Qian (钱国成) is currently a first-year Ph.D. student in Computer Science at KAUST. He is part of the Image and Video Understanding Lab (IVUL) advised by Prof. Bernard Ghanem. Guocheng received his B.Eng degree with first class honors from Xi’an Jiaotong University (XJTU), China in 2018.

“I want to make deep learning less costly and more efficient”, says Guocheng. Guocheng’s research interests lie in efficient deep learning. He currently works on Neural Architecture Search (reduce design cost), self-supervised learning (reduce labeling cost), and efficient neural network (reduce application/deployment cost).

Education

MSc in CS (3.95/4.0)
KAUST , 2019 - 2020
B.Eng in ME (3.93/4.3)
XJTU , 2014 - 2018
Exchange Program
HKUST , 2016

Interests with Paper

Neural Architecture Search ( SGAS)
Efficient Neural Network
Graph Convolutional Network ( DeepGCNs, PU-GCN)
3D representation learning

News

Publications so far: 2 CVPR, 1 TPAMI, 1 NeurIPS, 2 Arxiv
2021/10/10 : First-authored paper ASSANet: An Anisotropic Separable Set Abstraction for Efficient Point Cloud Representation Learning gets accepted to NeurIPS’21 as a spotlight paper. code NEW!.
2021/03/07 : Co-first authored paper DeepGCNs: Making GCNs Go as Deep as CNNs get accpected to journal TPAMI’21 . We achieve SOTA on PPI and SOTA on ModleNet40. We also add experiments in PartNet. I implement the code.
2021/03/01 : First authored paper PUGCN get accepted to CVPR’21
2020/12/18 : I graduate as a Master in Computer Science! Congradulation to myself!!!
2020/12/01 : I write some random Deep Learning stuff on my blog Deep-Learning-PhD-Wiki. Keep updating!
2020/03/04 : Co-first authored paper SGAS (Sequential Greedy Architecture Search) get accepted to CVPR’20. See arxiv, project website and Code. I take care the GCN part.
2020/01/21 : I serve as Teaching Assistant for course: CS390D Deep Learning (2020 Spring)
2019/05/03 : First authored paper “Trinity of Pixel Enhancement: a Joint Solution for Demosaicking, Denoising and Super-Resolution” is available here arxiv. Release PixelShift200 dataset, a full color sampled dataset for demosaicking related research.
2018/07/10 : I graduate from Xi’an Jiaotong University with the highest undergraduate honor (with GPA 3.93/4.3).
2018/06/20 : I join Sensetime Research as a research intern, supervised by Jimmy S. Ren.

Selected Publications

ASSANet: An Anisotropic Separable Set Abstraction forEfficient Point Cloud Representation Learning

Guocheng Qian, Hasan Abed Al Kader Hammoud, Guohao Li, Ali Thabet, Bernard Ghanem

October 2021 NeurIPS’21

We propose a novel Anisotropic Separable Set Abstraction module that makes PointNet++ faster and more accurate. We denote our modified PointNet++ as ASSANet, whose scaled version sets a new state-of-the-art on S3DIS with much faster speed.

Preprint Code