You can find my Google Scholar profile here.

  Journal articles

  Conference papers

Code & Data

  Software packages

  • TensorLy - Tensor operators and decompositions in Python

    Tensor methods are gaining increasing traction in machine learning. However, there are scant to no resources available to perform tensor learning and decomposition in Python. To answer this need we developed TensorLy. TensorLy is a state of the art general purpose library for tensor learning. Written in Python, it aims at following the same standard adopted by the main projects of the Python scientific community and fully integrating with these. It allows for fast and straightforward tensor decomposition and learning and comes with exhaustive tests, thorough documentation and minimal dependencies. It can be easily extended and its BSD licence makes it suitable for both academic and commercial applications.

    Download Paper
  • Discriminant Incoherent Component Analysis (Matlab)

    This software package contains a Matlab implementation of the solver introduced in Discriminant Incoherent Component Analysis, IEEE Transactions on Image Processing, 2016, for recovering class-specific incoherent components from (visual) data contaminated by noise of large magnitude.

    Download Paper
  • Sparse Low-Rank Representation (Matlab)

    This software package contains a Matlab implementation of the solver introduced in Music genre classification via joint sparse low-rank representation of audio features IEEE/ACM Trans. Audio, Speech, and Language Processing, 2014 for recovering a joint sparse low-rank representation from union of subspaces.

    Download Paper


  • SEWA Database

    The SEWA Database contains recordings of 408 people watching various advertisements and discussing them via a video-chat software. The recorded subjects are from six different cultural backgrounds: British, German, Hungarian, Greek, Serbian, and Chinese, 50-50 male-female ratio, with at least 6 subjects per culture in each age group 20+, 30+, 40+, 50+, 60+. The dyadic interactions are always in the native language and between people who know each other well or very well (including 70+ couples married for over 20 years). All dyadic interactions are transcribed.
    The annotations include: facial landmark tracking, AUs (AU1, AU2, AU4, AU12, AU17), head gestures (nod and shake), hand gestures, mimicry, continous annotations of valance and arousal, and continuous annotations of liking. This is the largest and the richest DB of human conversational and emotional behaviour that has been released so far.

  • The Conflict Escalation Resolution (CONFER) Database

    The Conflict Escalation Resolution (CONFER) Database is a collection of audio-visual recordings of naturalistic interactions from televised political debates suitable for the investigation of conflict behavior in dyadic interactions. It contains 120 audio-visual episodes in Greek language extracted from a total of 27 TV broadcasts, with total duration of approximately 142 minutes and a total number of 54 subjects (43 male, 11 female). The dataset is split into 2 sets which consist of recordings containing interactions that involve two or three participants, respectively. All 120 clips have been annotated by 10 experts in terms of continuous conflict intensity.

    Download Paper
  • Sound of Pixels Dataset

    The Sound of Pixels Dataset consists of 20 manually annotated audio-visual recordings of sound sources such as talking faces or music instruments. The primary usage of the dataset is to evaluate the performance of sound source localization methods, in the presence of distracting motions and noise.

    Download Paper



i.panagakis[ at ]
y.panagakis[ at ]

ICL Office

Department of Computing,
Imperial College London,
180 Queen's Gate,
Room 317, Huxley Building,
London SW7 2AZ, UK

MDX Office

Computer Science Department,
Middlesex University,
The Burroughs, Hendon,
London NW4 4BT, UK

Social Media