PhD, Research Associate,

University of Glasgow

Pattern Recognition, Machine Learning, Social Signal Processing

Sir Alwyn Williams Building, G12 8RZ


+44 141 330 4933



My primary research interests are in the area of machine learning, computer vision, and machine perception of social behavior. Within these areas, I explored the benefits of employing feature selection strategies to improve prediction performance. Indeed, feature selection is the process of identifying the few most important variables to facilitate data visualization, reducing training and utilization times, and defying the curse of dimensionality to improve prediction performance of any classifier from SVMs to deep neural networks.


My current research focuses on integrating feature selection and deep learning solutions to tease out the structure of the elaborate code behind social interactions (Human-Human & Human-Robot), making it possible for machines to read and write human body language.


We are working on the analysis of dynamic scenes using deep learning techniques to efficiently detect the pose of multiple people. Long short-term memory networks are used to model socially relevant factors, such as posture and gestures, to infer whether the human is open to an interaction, or they are paying attention to the robot, and other sort of social signals that we use in our daily conversations.



  • Discrete time evolution process descriptor for shape analysis and matching. Melzi, S., Ovsjanikov, M., Roffo, G., Cristani, M. and Castellani, U. ACM Transactions on Graphics, (2018). [pdf][bibtex]
  • Infinite Latent Feature Selection: A Probabilistic Latent Graph-Based Ranking Approach. G. Roffo, S. Melzi, U. Castellani, and A. Vinciarelli. In Conf. IEEE International Conference on Computer Vision (ICCV 2017). [pdf][bibtex]
  • Infinite Feature Selection. G. Roffo, S. Melzi and M. Cristani. In Conf. IEEE International Conference on Computer Vision (ICCV 2015). [pdf][bibtex]
  • Online Feature Selection for Visual Tracking. G. Roffo, S. Melzi. In Conf. The British Machine Vision Conference (BMVC 2016). [pdf][bibtex]
  • The Visual Object Tracking VOT2016 Challenge Results. Joint Paper: G. Roffo, S. Melzi et Al. In Conf. IEEE European Conference on Computer Vision Workshops (ECCV 2016). [pdf][bibtex]
  • Ranking to Learn: Feature Ranking and Selection via Eigenvector Centrality. G. Roffo, S. Melzi. Springer Book Chapter: New Frontiers in Mining Complex Patterns, 2017. [pdf][bibtex]




nVIDIA GPU Grant 2017

March 15, 2017


NVIDIA GPU grants are intended to enable researchers to begin a new project and/or gain the preliminary results to support a larger proposal to other funding agencies (see GPU Grant Program).

MATLAB Central Coin 2016

April 10, 2017


Matlab FileExchange - Recognition for Outstanding Contributions (2016) in Feature Selection. The Feature Selection Library (FSLib) received more than 3,000 unique downloads in 2016, avg. ~300 downloads pcm (see FSLib online).

VOT 2016 Trackers

January 15, 2017


A library of roughly 40 trackers which is now publicly available from the VOT page: download source code



I'm used to post news and updates also on:

Fast Downloads

Copyright 2016-2019 @ All Rights Reserved