Ray Anyi Rao

Ph.D. Candidate, Multimedia Laboratory
The Chinese University of Hong Kong
Hong Kong, China
Email: anyirao [at] ie.cuhk.edu.hk

Short Bio

Ray Anyi Rao is a Ph.D. candidate at Multimedia Laboratory in the Chinese University of Hong Kong, advised by Dahua Lin and Bolei Zhou. He also works closely with Sanja Fidler, UofT & Vector Institute. He is interested in computer vision, multimodality, video semantic understanding, cinematic style analysis and editing.
He got the B.S. from EE Department Honor Class, Nanjing University in 2018, ranking 1/183. He was fortunate to work with Dejing Dou, Daniel Lowd at AIM Lab, and Francis Lau at HKU CS.

News

2020-07: MovieNet is online with an easy-to-use toolkit and a detailed tutorial as a part of OpenMMLab.

2020-07: Three papers accepted to ECCV 2020! All about movie research. They are Cinematic Style Analysis, Online Multi-modal Person Search, and a holistic dataset MovieNet (Spotlight).

2020-05: Our CVPR 2020 work Local-to-Global Multi-modal Movie Scene Segmentation will appear at LUV 2020 (15-min talk) and Sight and Sound 2020 (5-min talk).

2020-02: One paper is accepted to CVPR 2020! It proposes a new temporal video segmenting method based on a new dataset. It is going to help divide a long video into semantic continuous short videos and output a structural representation. And it also provides research opportunities towards story/plot understanding in long videos with a semantic unit.

2020-01: HotFlip is included in AllenNLP and TextAttack

Publication

[1] A Unified Framework for Shot Type Classification Based on Subject Centric Lens
Anyi Rao, Jiaze Wang, Linning Xu, Xuekun Jiang, Qingqiu Huang, Bolei Zhou, Dahua Lin
European Conference on Computer Vision (ECCV), 2020
Also appear at Video Turing Test 2020 (5-min talk)
[Paper] [Webpage]

[2] MovieNet: A Holistic Dataset for Movie Understanding
Qingqiu Huang, Yu Xiong, Anyi Rao, Jiaze Wang, Dahua Lin
European Conference on Computer Vision (ECCV), 2020 (Spotlight)
[Paper] [Webpage] [Supplements] [Code]

[3] Online Multi-modal Person Search in Videos
Jiayue Xia, Anyi Rao⁺, Linning Xu, Qingqiu Huang, Dahua Lin
European Conference on Computer Vision (ECCV), 2020
[Paper] [Webpage]

[4] A Local-to-Global Approach to Multi-modal Movie Scene Segmentation
Anyi Rao, Linning Xu, Yu Xiong, Guodong Xu, Qingqiu Huang, Bolei Zhou, Dahua Lin
In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020
Also appear at LUV 2020 (15-min talk) and Sight and Sound 2020 (5-min talk)
[Paper] [Webpage] [Code]

[5] HotFlip: White-Box Adversarial Examples for Text Classification
Javid Ebrahimi, Anyi Rao, Daniel Lowd, Dejing Dou
In Proceedings of Annual Meeting of the Association for Computational Linguistics (ACL), 2018
It is included in several open-source NLP research libraries AllenNLP, TextAttack and OpenAttack
[Paper] [Poster] [AllenNLP] [OpenAttack] [Code]

[6] Automatic Music Accompaniment
Anyi Rao, Francis Lau
In arXiv:1803.09033, 2018
[Paper] [Presentation] [Code]

Research Experience

Multi-modal Video Analysis and Understanding
April.2018-March.2020 MMLab, Hong Kong S.A.R.
Advisor: Prof. Dahua Lin (Director) and Prof. Bolei Zhou (Innovators Under 35)

Cinematic style analysis in videos, published one ECCV20.

Multi-modal story and plot understanding in movies, published one ECCV20.

Scene understanding in movies, published one ECCV20 and one CVPR20.

Robust Training with Word-level Adversity for NLP
Sept.2017-April.2018 Advanced Integration and Mining Lab (AIM), Eugene, OR, United States of America
Advisor: Prof. Dejing Dou (Director, Head of Baidu Big Data Lab) and Prof. Daniel Lowd (Director of Graduate Studies)

Proposed an efficient word-level gradient-based adversarial examples generation approach for training robust models.

Evaluated the method across a wide range of sentence-level classification tasks and the method using adversarial training achieved excellent performances on benchmarks.

This work on Adversarial examples for NLP was featured in an article in The Register.

Automatic Music Accompaniment Using Probabilistic Machine Learning
Jul.2017-Aug.2017 The University of Hong Kong, Hong Kong S.A.R.
Advisor: Prof. Francis Lau (Associate Dean)

Proposed a fast decoding algorithm to deal with performance errors and reduce computational complexity from O(n²) to O(n). It is able to work in real-time with practical length scores.

Constructed a comprehensive system and developed a parallel Hidden Markov Model for score following.

Developed a new free open-source Windows-based automatic music follower and accompanist.

Awards

Hong Kong PhD Fellowship	2018
National Scholarship awarded by the China Ministry of Education, the highest honor in China	2015
Provincial-Level Merit Student awarded by the Jiangsu Provincial Government, the highest honor in Jiangsu Province	2017
Nanjing University Top-Grade Scholarship, awarded to 12 students out of all undergraduate students at the university, the highest honor in the university	2018
SenseTime Scholarship, awarded to 30 students out of all AI major undergraduate students in China	2017
Zhenggang Scholarship, top 40 students in Nanjing University	2016
Zhenggang Jingying Scholarship	2017
Nanjing University People Scholarship	2016, 2017
Nanjing University Outstanding Student Leader Award	2015
Nanjing University Outstanding Student Award	2016
Nanjing University Top Volunteer Excellence Award	2015
World ranking 32nd in 2016 Calculus World Cup	2016
Meritorious winner prize in the 2016 National Mathematical Contest in Modeling	2016
Best paper in the 2014 University Electronics Design Contest	2014
Gold Medal in Invitational National Mathematical Olympiad	2013

Undergrad Academic and Courses

Click to expand

Academic

His GPA ranked No.1 in each semester during his undergraduate studies with an overall GPA: 3.96/4.00 and Rank: 1/183. He finished major curricula in 2 years and learned a bunch of online courses. [Whole]

Undergrad Research Beginning

Express Bill Handwritten Telephone Number Recognition
Feb.2017-Jul.2017 Learning And Mining from DatA Group, Nanjing University
Director: Prof. Zhi-Hua Zhou (ACM, AAAI, AAAS, IEEE Fellow, Academy of Europe Foreign Member)

Incorporated an NLP lexicon-based transcription into an RCNN-CTC network and improved its performance on standard benchmarks.

Accomplished 95% accuracy on each digit and 90% accuracy on a series of telephone numbers.

Real-time 3D Surface Reconstruction Using Lidar (Light Detection And Ranging)
Aug.2016-Sept.2017 Visual Sensing and Graphics Lab (VISG Lab), Nanjing University
Supervisor: Prof. Sidan Du (Director)

Proposed a novel line of sight algorithm to real-time reconstruct surface and achieved state-of-the-art results.

Employed a new surface lattice data structure in implicit surface update for memory efficiency.

Presented a real-time 3D reconstruction pipeline for large-scale Lidar point clouds.

Implemented parallel computation to update the implicit surface faster and Motion Estimation and Mapping to register point clouds. [Report] [Video]

Undergrad Course Projects

Computer Vision 3D Human Poses Estimation from a Single Image [Presentation]
Used sparse coding to reduce the ambiguities in the 3D pose estimation; Enforced human-portion constraints to get a minimization problem

Convex Optimization Road Car Flow and Its Effect on Roads [Report]
Constructed models to depict road circumstances and to predict car motion using Markov Decision Process and OD estimated matrix; Improved the models in terms of traffic load and vehicle type.

Probability and Stochastic Process Monte Carlo Method for Multidimensional Integrals [Report]

Machine Learning Construction of a Spam Filtering System

Microcomputers and Interface Techniques x86 Assembly Language Programming [Report]

Signal Processing Single-Photon Detector [Report] [Presentation]
Designed a 64-channel low-noise pre-amplifier using a symmetric structure, reduced noise by 100 times, drew an 8-layers circuit board.

Volunteer and Leadership Experience

Click to expand

Co-Founder of a Children Care Volunteer Program Sep.2015-Dec.2015
Co-founded a psychological consulting program to promote left-behind children's growth and education. Volunteered to teach left-behind children Math and English in a junior high school located in the remote, underdeveloped Xiushui county. Recognized as a key team leader in the successful Warm One Hundred Campaign, which raised money for left-behind children. Our group received an excellence award from the China Foundation for Poverty Alleviation.

Vice President of a Young Volunteers Association at Nanjing University Jun.2015-Jun.2016
Organized and participated in over 100 out-of-school and 20 in-school activities covering over 1000 volunteers. Our association received a volunteer association excellence award.

Research Intern Group Leader of CHANGJIANG ELEC. TECH. [Media Report] July.2015-Aug.2015

Campus Ambassador of HUAWEI Aug.2017-Dec.2017

Student Volunteer of International Conference on Computer Vision (ICCV) 2019 Dec.2019

Miscellaneous

Teaching Experience
Head TA, IERG 4160, Image and Video Processing, (undergraduate and graduate level), Fall 2019
TA, IERG 3180, Microcontrollers and Embedded Systems Laboratory, Spring 2020

Patents
A Video Classification Method, China, 202010694811.1
A Video Editing Method, China, 202010694551.1
An Image Processing Method and Related Products, China, 202010450801.3

Hobbies
Love: 🌊🥥🏄‍♂️✈️🎬
Travel: 🇨🇳🇭🇰🇺🇸🇯🇵🇲🇴🇬🇧🇸🇬🇰🇷🇦🇪🇳🇱🇧🇪🇩🇪🇫🇷
Undergrad: Nanjing University Student Choir Member (Joyful Snowflakes written by Chih-mo Hsu), The NJU EE Department Young Volunteers Association Vice President, Bronze medal at Nanjing University 55th Sports Meeting
Language: A bit of Japanese and Cantonese, Native Mandarin, Full proficiency English