Haiyan Wang

Hi, I am a 5th year Ph.D. student in the Media Lab, Dept. Electrical Engineering at The City College of New York, CUNY, advised by Professor Ying-Li Tian. My current research focuses on 3D perception and scene understanding. Prior to CUNY, I finished my Bachelor degree from Beijing University of Posts and Telecommunications.

Email  /  Google Scholar  /  LinkedIn  /  Github

News

  • [2022.03.02] Our paper PSMNet about multi-view indoor layout-pose estimation is accepted by CVPR 2022.

  • [2021.10.22] Invited as speaker on Wonderland AI Summit 2021.

  • [2021.06.01] Work with Sing Bing Kang at Zillow as a research intern on 3D room layout estimation till August 2021.

  • [2021.03.02] Our paper FESTA about 3D scene flow estimation is accepted by CVPR 2021 as Oral Presentation.

  • [2020.12.01] Two papers are accepted by WACV 2021.

  • [2020.06.01] Work with Dong Tian at InterDigital as a research intern on 3D scene flow estimation till August 2020.

  • [2019.07.01] Our paper is accepted by BMVC 2019 as Oral Presentation.

Research

My research mainly focus on the 2D/3D scene understanding and adopts deep learning methods to various down-stream tasks including detection, segmentation and scene flow estimation to get the perception of the surrounding world.

Sequential Point Clouds: A Survey
Haiyan Wang, Yingli Tian,
Pending  
project page / PDF / video / demo

We propose an extensive review of deep learning research on sequential point clouds.

PSMNet: Position-aware Stereo Merging Network for Room Layout Estimation
Haiyan Wang, Will Hutchcroft, Yuguang Li, Zhiqiang Wan, Ivaylo Boyadzhiev, Yingli Tian, Sing Bing Kang
CVPR, 2022  
project page / PDF / video / demo

We propose a novel joint layout-pose estimation network from multi-view panorama images.

FESTA: Flow Estimation via Spatial-Temporal Attention for Scene Point Clouds
Haiyan Wang, Jiahao Pang, Muhammad Asad Lodhi, Yingli Tian, Dong Tian
CVPR, 2021   (Oral Presentation)
project page / PDF / video / demo

We propose a new spatial-temporal attention mechanism to estimate 3D scene flow from point clouds.

Subsurface Pipes Detection Using DNN-based Back Projection on GPR Data
Jinglun Feng, Liang Yang, Haiyan Wang, Yingli Tian, Jizhong Xiao
WACV, 2021
project page / PDF / video / demo

In this paper, we present MigrationNet, a learning-based approach to detect and visualize subsurface objects.

Self-supervised 4D Spatio-temporal Feature Learning via Order Prediction of Sequential Point Cloud Clips
Haiyan Wang, Liang Yang, Xuejian Rong, Jinglun Feng, Yingli Tian
WACV, 2021
project page / PDF / video / demo

Self-supervised dynamic feature learning of sequential point cloud.

Towards Efficient 3D Point Cloud Scene Completion via Novel Depth View Synthesis
Haiyan Wang, Liang Yang, Xuejian Rong, Yingli Tian
ICPR, 2020
project page / PDF / video / demo

3D point cloud completion

Weakly Supervised Semantic Segmentation in 3D Graph-Structured Point Clouds of Wild Scenes
Haiyan Wang, Xuejian Rong, Liang Yang, Jinglun Feng, Yingli Tian
Under the review.
project page / PDF / video / demo

Designed a joint end-to-end 2D-3D deep architecture to compute hierarchical to apply only 2D supervision for 3D semantic point cloud segmentation of wild scenes. In conjuction with the proposed Visible-Net to solve the objects occlusion problem.

GPR-based Subsurface Object Detection and Reconstruction Using Random Motion and DepthNet
Jinglun Feng*, Liang Yang*, Haiyan Wang, Yifeng Song , Jizhong Xiao,
ICRA, 2020.
project page / PDF / video / demo

Implemented visual inertial fusion to estimate the pose of the GPR sensor and proposed an improved random motion migration method which eliminates the limitation of current GPR data collection procedure which requires the straight line motion along survey grid.

Towards Weakly Supervised Semantic Segmentation in 3D Graph-Structured Point Clouds of Wild Scenes
Haiyan Wang, Xuejian Rong, Liang Yang, Shuihua Wang, Yingli Tian
BMVC, 2019   (Oral Presentation, 3.8% acceptance rate)
project page / PDF / video / demo

Designed a joint 2D-3D deep architecture to compute hierarchical to apply only 2D supervision for 3D semantic point cloud segmentation of wild scenes. Integrated graph-based convolution and a novel reprojection method, named perspective rendering to enforce the 2D and 3D geometric correspondence.

Towards Accurate Instance-level Text Spotting With Guided Attention
Haiyan Wang, Xuejian Rong, Yingli Tian
ICME, 2019
PDF / Bibtex

Designed a deep learning framework for text detection from the instance-aware segmentation perspective. Specifically, a text-specific attention model and a global enhancement block are introduced to enrich the semantics of text detection features.

Teaching

TA, Spring Term, 2021:    I2200: Digital Image Processing (The City College of New York). Grade

TA, Spring Term, 2022:    I2200: Digital Image Processing (The City College of New York). Grade

Mentoring

Munib Ahsan (Oct 2020 - 2021 May):    Software Engineer at Northrop Grumman Corporation.

Zoya Shafique (Oct 2021 - ):    Master at CCNY.


Last update: 2022.03. Thanks.