Dr. Timothy Hospedales

Chen Change Loy

Post-doctoral Researcher

Vision Group
Electronic Engineering and Computer Science
Queen Mary University, London
E1 4NS
tmh at dcs dot qmul dot ac dot uk

I am a post-doc researcher with the vision group at QMUL. I coordinated the behaviour monitoring work package for the SAMURAI EU FP7 project at QMUL ('08-11). During my PhD, I worked with Sethu Vijayakumar in the Statistical Machine Learning and Motor Control group and Edinburgh's IPAB and ANC institutes.

My background spans informatics & neuroscience (PhD Neuroinformatics, Edinburgh, 2008; MSc Neuroinformatics, Edinburgh 2004), machine learning (MSc Informatics, Edinburgh, 2003) and computer science (BA Computer Science, University of Cambridge, 2002).

My academic interests include Bayesian probabilistic modelling and machine learning: applied variously to computer vision, behaviour understanding, data mining, data fusion, robotics, motion-capture, novel human-computer interfaces and theoretical neuroscience.

News

Stream-based Joint Exploration-Exploitation Active Learning, CVPR 2012
Learning from streams of evolving and unbounded data is an important problem, for example in visual surveillance or internet scale data. For such large and evolving real- world data, exhaustive supervision is impractical, particularly so when the full space of classes is not known in advance therefore joint class discovery (exploration) and boundary learning (exploitation) becomes critical. Active learning has shown promise in jointly optimising exploration-exploitation with minimal human supervision. However, existing active learning methods either rely on heuristic multi-criteria weighting or are limited to batch processing. In this paper, we present a new unified framework for joint exploration-exploitation active learning in streams without any heuristic weighting. Extensive evaluation on classification of various image and surveillance video datasets demonstrates the superiority of our framework over existing methods.
PDF

Unsegmented sports news dataset available (ICDM'11). Includes reference extracted STIP features and ground-truth multi-label annotation.

Sample code for MCTM (ICCV'09, IJCV'11) and WSJTM (ACCV'10, PAMI'11)

About Me

Education

  1. Ph.D. Neuroinformatics, University of Edinburgh, 2008
    Thesis: Bayesian Multisensory Perception
    Advisor: Sethu Vijayakumar
    Examiners: Carl Rasmussen and Amos Storkey

  2. MSc, Neuroinformatics, University of Edinburgh, 2004
    Thesis: Functional Organization & Plasticity in the Vestibulo-Ocular Reflex
    Advisor: Mayank Dutia

    3. MSc, Informatics, University of Edinburgh, 2003
    Thesis: Optimal Learning in Neurons with Active Dendritic Spines
    Advisor: David Barber

  3. BA, Computer Science, University of Cambridge, 2002
    Dissertation: A Probabilistic Approach for Commodity Eye-Tracking
    * Best Dissertation Prize, University of Cambridge Computer Laboratory
    * UK SET Computer Science Student of the Year Award
    Advisor: David Mackay

Professional Experience

  1. Vision Semantics Limited, United Kingdom
    Consultant, 2009-Present
    Projects: Super Resolution, Egocentric Behaviour Profiling

  2. Queen Mary University of London, United Kingdom
    Postdoctoral Researcher, 2008-Present
    Projects: SAMURAI

Academic and Services

    Conference Reviewer

    • ICCV, CISP, ICNC-FSKDD

    Journal Reviewer

    • IEEE Transactions on Image Processing
    • Neurocomputing
    • Journal of Statistical Planning and Inference
    • SPIE Optical Engineering

Publications

Visit my Google Scholar Profile.

Active Learning

Behaviour Understanding

General Computer Vision and Machine Learning

Neuroscience

Journal

Conference

Bookchapter / Workshop / Thesis

2012

2011

2010

2009

2008

2007

Download datasets, code

Datasets

Unsegmented Sports News

Unsegmented Sports News Videos.

Sports newscasts collected from youtube.
74 videos, Xvid codec, 240x320 pixels, 60-320 sec each.
10 classes: Swimming, American Football, Soccer, Tennis, Volley Ball, Track, Baseball, Golf, Basketball, Wrestling.
Average 3 sports/classes per video.

Tasks: Multi-label/Weakly supervised learning, Segmentation, Activity Detection / Classification.
Publications: ICDM'11.

Download Raw Data (1GB) Download Extracted Feats & Annotations (3MB)

QMUL Junction Dataset

A busy traffic intersection.

Video length: 1 hour (90000 frames)
Frame size: 360x288
Frame rate: 25 Hz
Compression codec: ffdshow mpeg-4

Tasks: Behaviour Profiling, Anomaly Detection, Background Subtraction.
Publications: ICCV'09, IJCV'11, PAMI'11.

Details and download ...

Code

MCTM: Markov Clustering Topic Model

Learn a temporally correlated hierarchical topic model.
Download

Reference:

WSJTM: Weakly Supervised Joint Topic Model

Learn a topic model for rare and subtle behaviours.
Download

Reference: