Jianfeng Zhang (张健锋)

Undergraduate Student

Vision & Machine Learning Lab,
Block E4 #08-24,
4 Engineering Drive 3,
National University of Singapore

Email: jf.zhang958 dot outlook dot com


I am currently a research assistant at NUS Vision & Machine Learning Lab, advised by Prof. Jiashi Feng and working closely with Dr. Xuecheng Nie. Previously, I received the B.S. degree from Department of Applied Mathematics in Wuhan University, supervised by Prof. Xiaoping Zhang for undergraduate research. During my undergrad study, I was fortunate to be an intern at Farsee2 and OrionStar Computer Vision Lab.

My research interests include 2D/3D Vision, Deep Learning and their applications.


Single-stage Multi-person Pose Machines
Xuecheng Nie, Jianfeng Zhang, Shuicheng Yan, Jiashi Feng
IEEE International Conference on Computer Vision (ICCV), 2019


Interactive Binary Image Segmentation with Edge Preservation
Jianfeng Zhang, Liezhuo Zhang, Yuankai Teng, Xiaoping Zhang, Song Wang, Lili Ju
Arxiv Preprint


Developing a LSTM based Model for Predicting Water Table Depth
Jianfeng Zhang, Yan Zhu, Xiaoping Zhang, Ming Ye, Jinzhong Yang
Journal of Hydrology (JH), 2018

[pdf] [code]


Deeplab V3+ in PyTorch

We reimplement Deeplab V3+ in PyTorch, and evaluate it on Pascal VOC 2012 and Cityscapes datasets. We apply Deeplab V3+ to extract the expected object from multi-view images for stereo matching, in order to get better 3D reconstruction results.


Deep Grabcut (DeepGC)

We replicate "Deep Grabcut for Object Selection", this paper proposes a segmentation approach that uses a rectangle as a soft constraint by transforming it into an Euclidean distance map. A convolutional encoder-decoder network is trained end-to-end by concatenating images with these distance maps as inputs and predicting the object masks as outputs. We implement this paper using PyTorch and train this model using Pascal VOC 2012 and SBD datasets.


Video Activity Recognition in PyTorch

We reimplement several video activity recognition models including C3D, R2Plus1D, R3D in PyTorch, and evaluate these models on HMDB51 and UCF101 datasets. We also apply C3D model into an online web game demo called "You Perform, I Guess!". This demo obtains Excellent Demo Award in DeeCamp 2018.

[code] [demo]

Interactive Segmentation on RGBD Image

We replicate “Interactive Segmentation on RGBD Images via Cue Selection”, this paper proposes a novel interactive segmentation algorithm which can incorporate multiple feature cues like color, depth and normals in a graph cut framework.

[paper] [code]

Total Variation Image Segmentation Model Based on Primal-dual Method

We implement total variation image segmentation algorithm and applied primal-dual method to optimize it. In order to obtain better results, we integrate user provided information into algorithm. In addition, we also combine the model with K-means to improve the final results.

[pdf] [code]

Video Object Segmentation Framework

We build a Video Object Segmentation Framework. In this framework, we combine MobileNetV2, YOLOV2 and Deep Grabcut to do video object segmentation.
End to End Network Image Text Detection and Recognition

We build a faster-rcnn like model for end to end network image text detection and recognition. Our method achieves 0.634 accuracy and ranks 13/1488 on ICPR MTWI 2018 Challenge on Tianchi.
Clothing Fashion Style Recognition

We build our fashion style recognition model based on ResNet, DenseNet and VGG. We also use Deeplab V3+ to extract clothes and remove background, which could help to denoise. In addition, we apply model ensemble method to improve our results. Finally, our method achieves 0.610 F2 score and ranks 9/213 on JD AI Fashion Challenge.

Honors & Awards

Excellent Demo Award, DeeCamp, 2018
Best Bachelor Thesis Award, School of Mathematics and Stastics, Wuhan University, 2018
Award for Merit Student of Wuhan University, 2015-2017


2017-2018SpringC Language Programming
2017-2018FallData Structures and Algorithms in Python

© Jianfeng Zhang | Last updated: 24/07/2019