top of page
Publications 

RESEARCH

Occlusion-aware Video Temporal Consistency

Chun-Han Yao, Chia-Yang Chang, and Shao-Yi Chien

ACM Multimedia, 2017

Paper

For image processing techniques like color transfer, HDR tone mapping, white balance, and dehazing, naively  applying them to videos frame-by-frame often leads to temporal inconsistency, which is especially severe when occlusion occurs. We solve it with a framework of keyframe selection, temporal warping, content compensation, and post-processing.

Seamless mmWave 5G Design: Outage Reduction with Joint Scheduling and Power Allocation

Chun-Han Yao, Yin-Yi Chen, and Hung-Yu Wei

PIMRC workshop, 2017

Paper​

In mmWave communications, signal is easily blocked by obstacles, and user mobility is likely to cause a misalingment between transmitter and receiver beams, both result in unpredictable channel conditions. In this project we propose a joint scheduling and power allocation algorithm to minimize outage probability caused by user mobility.

Progressive Domain Adaptation for Object Detection

Han-Kai Hsu, Wei-Chih Hung, Hung-Yu Tseng, Chun-Han Yao, Yi-Hsuan Tsai, Maneesh Singh, Ming-Hsuan Yang

Computer Vision and Pattern Recognition (CVPR) Workshops, 2019

Paper

Progressive Domain Adaptation for Object Detection

Han-Kai Hsu, Chun-Han YaoYi-Hsuan Tsai, Wei-Chih Hung, Hung-Yu Tseng, Maneesh Singh, Ming-Hsuan Yang

Winter Conference on Applications of Computer Vision (WACV), 2020

Paper

Video Object Detection via Object-level Temporal Aggregation

Chun-Han YaoChen Fang, Xiaohui Shen, Yangyue Wan, Ming-Hsuan Yang

European Conference on Computer Vision (ECCV), 2020

Paper

Discovering 3D Parts from Image Collections

Chun-Han Yao, Wei-Chih Hung, Varun Jampani, Ming-Hsuan Yang

International Conference on Computer Vision (ICCV), 2021

Paper   Supp   Page   Video

Millimeter-wave Multi-hop Wireless Backhauling for 5G Cellular Networks

B. P. S. Sahoo, Chun-Han Yao, and Hung-Yu Wei

VTC Spring, 2017

Paper​

In this project we propose an inter-BS scheduling algorithm for mmWave multi-hop wireless backhauling, which seeks to optimize the system throughput via dynamic allocation of tranmission periods according to data load and link quality.

Federated Multi-Target Domain Adaptation

Chun-Han Yao, Boqing Gong, Yin Cui, Hang Qi, Yukun Zhu, Ming-Hsuan Yang

Winter Conference on Applications of Computer Vision (WACV), 2022

Paper   ArXiv   Supp

Learning Visibility for Robust Dense Human Body Estimation (VisDB)

Chun-Han Yao, Jimei Yang, Duygu Ceylan, Yi Zhou, Yang Zhou Ming-Hsuan Yang

European Conference on Computer Vision (ECCV), 2022

ArXiv   Supp   Page

LASSIE: Learning Articulated Shapes from Sparse Image Ensemble

Chun-Han Yao, Wei-Chih Hung, Yuanzhen Li, Michael Rubinstein, Ming-Hsuan Yang, Varun Jampani

Neural Information Processing Systems (NeurIPS), 2022

ArXiv  Supp   Page   Video

Hi-LASSIE: High-Fidelity Articulated Shape and Skeleton Discovery from Sparse Image Ensemble

Chun-Han Yao, Wei-Chih Hung, Yuanzhen Li, Michael Rubinstein, Ming-Hsuan Yang, Varun Jampani

CVPR, 2023

ArXiv  Page   Video

Example-based Video Color Transfer

Chun-Han Yao, Chia-Yang Chang, and Shao-Yi Chien

ICME, 2016

Paper        Video

Color transfer is an image processing technique commonly used to fix images with undesirable colors, enhance lighting conditions, or produce a special style to express specific emotion. In this paper we propose an efficient algorithm for example-based image/video color transfer and a quality metric for video temporal consistency.

Projects
train.png

2018/03 - 2018/08

“3D Context Encoder for Defect Detection”,

3D Vision Lab, UCSD, instructed by Prof. Manmohan Chandraker

In this work we propose an unsupervised learning framework to extract contextual feature of 3D models, which is shown capable of reconstructing arbitrarily masked parts of a 3D scanned model and detect the defects in it.

螢幕快照 2018-08-11 上午11.33.41.png

2018/03 - 2018/06

“Imitation Learning from Suboptimal Demonstrations”,

Reinforcement Learning for Robotics, UCSD, instructed by Prof. Michael Yip

Report​

In this work we propose multiple approaches to integrate expert demos with reinforcement learning (RL) algorithms. Experimental results on OpenAI Gym environments show that RL agents learn much more efficiently via the proposed methods even with sub-optimal demonstrations.

螢幕快照 2018-08-11 上午11.32.24.png

2018/01 - 2018/03

“Find the Nuclei: Biomedical Image Segmentation”,

Neural Networks for Pattern Recognition, UCSD, instructed by Prof. Garrison Cottrell

Report​

Identifying the nuclei of cells in diverse medical images is a crucial yet tedious task for medical analysis. We develop a framework with U-Net architecture, boundary-aware objective function, and Gaussian Mixture Model for pixel clustering to automate the process. The proposed approach achieves a mean Average Precision of 0.413, which is top 10% in Kaggle 2018 Data Science Bowl.

螢幕快照 2018-08-11 下午12.25.45.png

2017/09 - 2017/12

“3D Human Pose Reconstruction - A Convex Relaxation Approach”,

Convex Optimization, UCSD, instructed by Prof. Chung-Kuan Chen

Report​

In this project we reconstruct 3D human pose from single 2D image. Given a set of basis 3D shapes, we formulate the problem into a convex optimization and solve it with Alternating Direction Method of Multipliers (ADMM).

2016/02 - 2016/06

“Simple Gesture Recognition Using Wireless Signal”,

Communication System Lab, NTU, instructed by Prof. I-Hsiang Wang

Report​

As smart home applications are gaining popularity, the demand for human movement sensing and gesture recognition is rising accordingly. In this project a simple gesture recognition system using wireless signal is implemented by USRP and LabVIEW.

2015/09 - 2016/01

“Predicting Facebook Post Likes”,

Data Science, NTU, instructed by Prof. Ming-Syan Chen

Report

In this project we seek to find the factors in a post that attract more likes. Hence, features of each recent post from certain Facebook pages are collected, and a model combining NLP, PCA, Naive Bayes, and desision tree, is developed, which is able to accurately predict the number of post likes.

2015/09 - 2016/01

“Emotional Impact of Social Media Activity”,

Psychoinformatics & Neuroinformatics, NTU, instructed by Prof. Tsung-Ren Huang

In this project we develop a mobile application for time-lag studies and survey on the emotional impact of social media activity. As a result, direct interaction including messaging and commenting on posts have a positive effect on people's emotion.

2015/02 - 2015/06

“Vehicle and Pedestrian Detection for Intelligent Traffic System”,

Digital Circuit Lab, NTU, instructed by Prof. Shao-Yi Chien

Report        Video

In intelligent traffic systems, vehicle and pedestrian detection is the key to providing useful information and reliable service.  Consequently, we develop a vehicle and pedestrian detection system connecting cameras and  FPGA boards for hardware setup in such systems.

2015/02 - 2015/06

“Directional Traffic Light Base Station”,

Introduction to Wireless & Mobile Networking, NTU, instructed by Prof. Hung-Yu Wei

Paper

Traffic light base station is a promising solution to intelligent traffic systems. We simulate different scenarios of traffic light base stations using directional mmWave communications and show the pros and cons of the proposed model.

2015/02 - 2015/06

“Auto-watering System Using Wastewater from Household Appliance”,

Internet of Things, HackNTU

As an IoT application, we implement an automatic watering system using wastewater provided by household appliance.The system componenents include a LinkIt ONE board, an Arduino board, a soil moisture sensor, and a water level sensor.

2014/09 - 2015/01

“Heart Rate Monitoring During Physical Exercise Using Wrist-type PPG Signals”,

2015 Signal Processing Cup, IEEE, advised by Prof. Jian-Jiun Ding

Paper

In this project we propose a real-time algorithm to moniter the heart beat rate using PPG signals provided by wrist-type devices. By employing signal processing techniques such as PCA, ICA, Hilbert -Huang transformation, time-frequency analysis, etc, we can track the heart rate with high accuracy.

bottom of page