I'm Fangqiao Hu (胡芳侨), a third year PhD candidate advised by Prof. Hui Li in Center of Structural Monitoring and Control, Harbin Institute of Technology (HIT). I got my Bachelor's degree from HIT, too.
My research topics are 3D computer vision and machine learning and their applications in structural health monitoring and civil engineering, for example, 3D reconstruction for civil structures to enable remote visual inspection.
I attended the 7th World Conference on Structural Control and Monitoring (7WCSCM) held in Qingdao, China, and 2018 Asia-Pacific-Euro Summer School on Smart Structures Technology (APESS 2018) held in Qingdao and Harbin, China. I also worked as a volunteer there.
I attended the Chinese Conference on Pattern Recognition and Computer Vision (PRCV 2018) held in Guangzhou, China.
I attended the Conference on Computer Vision and Pattern Recognition (CVPR 2019) held in Long Beach, California, U.S.
I attended the Engineering Mechanics Institute Conference (EMI 2019) held in Caltech, Pasadena, California, U.S. I also gave a presentation (Slides) at EMI 2019.
I took part in the International Project Competition for Structural Health Monitoring (IPC SHM 2020). We won the 1st prize (official results) out of 31 groups. The code is released.
A simplified implementation of Projective Dynamics (SIGGRAPH 2014).
[Code]
Bachelor's thesis, 2018. Damage detection based on 2D images and 3D point clouds.
[Paper]
This project was done in the APESS 2018 summer school, a steel girder crack detection network. It improves the performance on crack detection by adding deformable CNN modules.
[Code]
It provides a single 3D object segmentation network on point cloud, realized by predicting a 3D orientated bounding box.
[Code]
The recursive BiTreeNet provides a recursive binary tree decoder to hierarchically learn graph layouts and contents of nodes.
[Code]
A single SASA neuron can solve the linearly inseparable “exclusive or” (XOR) problem, which requires at least two hidden layers for a conventional neural network.
[Code]
Fangqiao Hu, Jin Zhao, Yong Huang, Hui Li*
A structure-aware learning-based cable-stayed bridge 3D reconstruction framework is proposed. The encoder part of the network uses both multiview images and a photogrammetric point cloud as input, whereas the decoder part uses a recursive binary tree network to model a high-level structural relation graph and low- level 3D geometric shapes.
(Computer-Aided Civil and Infrastructure Engineering, IF=8.552)
Jin Zhao†, Fangqiao Hu†, Weidong Qiao, Weida Zhai, Yang Xu, Yuequan Bao, Hui Li* (†: Equal contribution)
Despite impressive breakthroughs in deep learning and computer vision, the pixel-wise identification of tiny objects such as minor cracks in high-resolution images with complex background disturbances still remains challenging. To solve the problem, this study first proposes a novel Self-Attention-Self-Adaption (SASA) neuron computing model as follows: the Self-Attention module helps the neuron focus on the most significant receptive fields by softmax score ranking when processing large-scale feature maps, and the Self-Adaption module enables deeper feature extraction using only one single neuron which includes a sub-net of multilayer perceptron.
(Smart Structures and Systems, IF=3.557)
[Paper] [Code]
Jin Zhao, Fangqiao Hu, Yang Xu, Wangmeng Zuo, Jiwei Zhong, Hui Li*
This paper proposes a novel framework to recognize 3D pose and dense vibration displacement of a structure with known size using monocular videos. The 3D pose can obtain the structural displacement of invisible elements in videos. The proposed framework consists of two consecutive deep learning modules, i.e., CompNet and ParaNet, to provide semantic image segmentation and extract pose parameters, respectively.
(Computer-Aided Civil and Infrastructure Engineering, IF=8.552)
[Paper] [Code]