Loris Bazzani

I am currently (2016-) a Computer Vision Scientist at Amazon in Berlin, Germany.

I obtained my Ph.D. in Computer Science from the University of Verona (Italy) in 2012 supervised by Prof. Vittorio Murino and Prof. Marco Cristani. During my Ph.D., I spent 6 amazing months at the University of British Columbia supervised by Prof. Nando de Freitas. Before the current position, I was a postdoctoral fellow (2014-2015) at Dartmouth College working with Prof. Lorenzo Torresani and I was a postdoctoral fellow (2012-2013) at the Italian Institute of Technology working with Prof. Vittorio Murino.


Curriculum Vitae / Google scholar / Linkedin / Github / some cool videos

Sections: unpublished work / journal papers / conference papers / book chapter

Research

Learning is a continuous, potentially unlimited and complex process, hard to replicate with machines. The goal of my research is to investigate models and algorithms that learn from the visual world (i.e., images and videos).

I spent my Ph.D. working on probabilistic models that deal with the temporal dimension in computer vision problems from both the engineering point of view (tracking and social interaction models) and the cognitive point of view (attentional models). I also worked on person re-identification, interesting instance recognition problem in video-surveillance.

My research is focused on object localization, detection and segmentation in images and videos as well as attentional models for computer vision.

News

2016
- Apr 5: Paper on Attentional Modeling is out in ARXIV.
- Mar 15: Paper on Covariance Descriptors accepted at CVPR 2016.
- Jan 29: Paper on Object Localization with deep nets accepted at WACV 2016.
- Jan 10: I am excited to join Amazon in Berlin, Germany.
2015
- Oct 26: Guest editor for the CVIU Image and Video Understanding in Big Data special issue.
- Oct 25: Paper on Multi-view Learning for classification accepted at JMLR.
- Sep 7: Reviewer for CVPR 2016.
- Aug 8: I am on the job market. Contact me if interested.
- Jun 3: Code of Self-taught Object Localization available on github.
- Mar 18: Updated paper of Multi-view Learning available on Arxiv.
- Jan 20: Web Chair of the workshop GROW@CVPR 2015.

Project pages

  • Object recognition and localization:
    • STL: self-taught object localization with deep nets
    • MVL: multi-view learning for recognition
  • Attentional modeling for vision:
    • Attentional RBM tracker: a tracker that simulates the human attention mechanism
    • RMDN: a recurrent mixture density network for saliency prediction
  • Person re-identification:
  • Social group detection and analysis:
    • IRPM: subjective view frustum for group detection
    • FM dataset for group detection and tracking

Recent Unpublished Work

Recurrent Mixture Density Network for Spatiotemporal Visual Attention
L. Bazzani, H. Larochelle, L. Torresani
Arxiv, 2016
Project page / arXiv / video / bibtex
@article{Bazzani:Arxiv2016,
  title={Recurrent Mixture Density Network for Spatiotemporal Visual Attention},
  author={Bazzani, Loris and Larochelle, Hugo and Torresani, Lorenzo},
  journal={arXiv preprint arXiv:1603.08199},
  year={2016}
}
  
We propose an attentional model that learns where to look in a video directly from human fixation data. The model is a combination of 3D ConvNet, RNN and mixture density network which models the saliency map.

Journal papers

A Unifying Framework in Vector-valued Reproducing Kernel Hilbert Spaces for Manifold Regularization and Co-Regularized Multi-view Learning
H. Q. Minh, L. Bazzani, V. Murino
Journal of Machine Learning Research (JMLR), 2016
MVL code / arXiv / bibtex
@article{Minh:JMLR16,
  author  = {H{{\`a}} Quang Minh and Loris Bazzani and Vittorio Murino},
  title   = {A Unifying Framework in Vector-valued Reproducing Kernel Hilbert Spaces 
  for Manifold Regularization and Co-Regularized Multi-view Learning},
  journal = {Journal of Machine Learning Research},
  year    = {2016},
  volume  = {17},
  number  = {25},
  pages   = {1-72},
  url     = {http://jmlr.org/papers/v17/14-036.html}
}
    
Extension of our ICML 2013 paper. We propose a new formulation that includes the multi-view SVM algorithm and the optimization of the combination operator.
Joint Individual-Group Modeling for Tracking
L. Bazzani*, M. Zanotto*, M. Cristani, V. Murino
IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2015
FM dataset / video / supplement / bibtex
@article{Bazzani:PAMI15,
  author={Bazzani, L. and Zanotto, M. and Cristani, M. and Murino, V.},
  journal={Pattern Analysis and Machine Intelligence,
           IEEE Transactions on},
  title={Joint Individual-Group Modeling for Tracking},
  year={2015},
  month={April},
  volume={37},
  number={4},
  pages={746-759},
  doi={10.1109/TPAMI.2014.2353641},
  ISSN={0162-8828}
}
    
The presented model is a combination of the tracking model prensented at CVPR 2012 and the detection model presented at BMVC 2012
Symmetry-driven accumulation of local features for human characterization and re-identification
L. Bazzani, M. Cristani, V. Murino
Computer Vision and Image Understanding (CVIU), 2013.
SDALF code / bibtex
@article{Bazzani:CVIU13,
  title = {Symmetry-driven accumulation of local features
           for human characterization and re-identification},
  author = {Bazzani, Loris and Cristani, Marco and Murino, Vittorio},
  journal = {Comput. Vis. Image Underst.},
  year = {2013},
  month = feb,
  number = {2},
  pages = {130--144},
  volume = {117},
  doi = {10.1016/j.cviu.2012.10.008},
  issn = {1077-3142},
  issue_date = {February, 2013},
  numpages = {15},
  owner = {lbazzani},
  publisher = {Elsevier Science Inc.}
}
    
This paper is an extension of SDALF (CVPR 2010 paper) to be used as appearance model for tracking
Social interactions by visual focus of attention in a three-dimensional environment
L. Bazzani, D. Tosato, M. Cristani, M. Farenzena, G. Pagetti, G. Menegaz, and V. Murino
Expert Systems 2013
IRPM code / video / bibtex
@article{Bazzani:ExpSys13,
  title = {Social Interactions by Visual Focus of Attention
           in a Three-Dimensional Environment},
  author = {Bazzani, L. and Tosato, D. and Cristani, M. and
            Farenzena, M. and Pagetti, G. and Menegaz, G. and
            Murino, V.},
  journal = {Expert Systems},
  year = {2013}
}
    
This is an extension of the work presented at PRAI*HBA 2009 workshop and ICIAP 2009 in which we studied how to detect interactions between individuals
Learning where to attend with deep architectures for image tracking
M. Denil, L. Bazzani, H. Larochelle, and N. de Freitas
Neural Computation, 2012
RBM tracker code / video / bibtex
@article{Misha:2012NECO,
  title = {Learning where to Attend with Deep Architectures
           for Image Tracking},
  author = {Denil, M. and Bazzani, L. and Larochelle, H.
            and {de Freitas}, N.},
  journal = {Neural Computation},
  year = {2012}
}
    
We extended our ICML 2011 paper including a Bayesian optimization technique that explores parts of the image to automatically find the best region for tracking
Multiple-shot person re-identification by chromatic and epitomic analyses
L. Bazzani, M. Cristani, A. Perina, and V. Murino
Pattern Recognition Letters (PRL), 2012
CAVIAR4REID dataset / bibtex
@article{Bazzani:PRL11,
  title = {Multiple-shot person re-identification by chromatic
           and epitomic analyses},
  author = {Bazzani, L. and Cristani, M. and Perina, A. and
            Murino, V.},
  journal = {Pattern Recognition Letters},
  year = {2012},
  doi = {10.1016/j.patrec.2011.11.016},
  issn = {0167-8655}
}
    
This is an extension of our ICPR 2012 with the introduction of SDALF asymmetry axes and a deeper experimental analysis.
Invited paper since I obtained the ICPR 2012 IBM Best Student Paper

Selected Conference papers

(see my cv for the full list of papers)
Approximate Log-Hilbert-Schmidt distances between covariance operators for image classification
H. Q. Minh, M. San Biagio, L. Bazzani, V. Murino
In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016
bibtex
@conference{Minh:CVPR16
  title     = {Approximate Log-Hilbert-Schmidt distances
between covariance operators for image classification},
  author    = {Minh, H. Q. and San Biagio, M. and Bazzani, L. and Murino, V.},
  booktitle   = {In IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2016}
}
  
This paper presents an object recognition framework using infinite-dimensional covariance operators. We provide an approximation for the Log-Hilbert-Schmidt distance between covariance operators that is efficient and scalable.
Self-taught object localization with deep networks
L. Bazzani, A. Bergamo, D. Anguelov, L. Torresani
In IEEE Winter Conference on Applications of Computer Vision (WACV), 2016
STL code / arXiv / bibtex
@conference{Bazzani:WACV16
  title     = {Self-taught Object Localization with Deep Networks},
  author    = {Bazzani, L. and Bergamo A. and Anguelov, D. and
               Torresani, L.},
  booktitle   = {In IEEE Winter Conference on Applications of Computer Vision (WACV)},
  year      = {2016}
}
  
We leverage deep convolutional networks trained for whole-image recognition to localize objects in images without additional human supervision.
Weighted bag of visual words for object recognition
L. Bazzani*, M. San Biagio*, M. Cristani, V. Murino
In IEEE International Conference on Image Processing (ICIP), 2014
bibtex
@conference{SanBiagio:ICIP14,
  title = {Weighted bag of visual words for object recognition},
  author = {San Biagio, Marco and Bazzani, L. and
            Cristani, M. and Murino, V.},
  booktitle = {In IEEE International Conference on Image
               Processing (ICIP)},
  year = {2014}
}
   
We propose a weighted version of bag of feature where the weight is computed accordingly to the patch salience.
A unifying framework for vector-valued manifold regularization and multi-view learning
H. Q. Minh, L. Bazzani, V. Murino
The 30th International Conference on Machine Learning (ICML), 2013
MVL code / bibtex
@inproceedings{Minh:ICML13,
  title = {A unifying framework for vector-valued manifold
           regularization and multi-view learning},
  author = {Minh, H. Q. and Bazzani, L. and Murino, V.},
  booktitle = {Proceedings of the 30th International Conference
               on Machine Learning (ICML-13)},
  year = {2013},
  editor = {Sanjoy Dasgupta and David Mcallester},
  month = may,
  number = {2},
  pages = {100-108},
  publisher = {JMLR Workshop and Conference Proceedings},
  volume = {28}
}
    
The multi-view learning model performs multi-modal/multi-feature object classification in the semi-supervised setup
Semi-supervised multi-feature learning for person re-identification
D. Figueira, L. Bazzani, H.Q. Minh, M. Cristani, A. Bernardino, V. Murino
In International Conference on Advanced Video and Signal-based Surveillance (AVSS), 2013
bibtex
@INPROCEEDINGS{Dario:AVSS2013,
  author={Figueira, D. and Bazzani, L. and Ha Quang Minh and
          Cristani, M. and Bernardino, A. and Murino, V.},
  booktitle={In International Conference on Advanced Video and
          Signal Based Surveillance (AVSS)},
  title={Semi-supervised multi-feature learning for person
          re-identification},
  year={2013},
  month={Aug},
  pages={111-116},
  doi={10.1109/AVSS.2013.6636625}
}
    
Application paper of our multi-view learning model (ICML 2013) in the context of person re-identification
Person re-identification with a PTZ camera: an introductory study
P. Salvagnini, L. Bazzani, M. Cristani, V. Murino
In International Conference on Image Processing (ICIP), 2013
bibtex
@inproceedings{Salvagnini:ICIP2013,
  title={Person re-identification with a PTZ camera: An introductory
          study.},
  author={Salvagnini, P. and Bazzani, L. and Cristani, M. and
          Murino, V.}
  booktitle = {In International Conference on Image
          Processing (ICIP)},
  year = {2013},
}
    
We demostrate the potential use of PTZ cameras to acquire higher resolution features in person re-identification
Decentralized particle filter for joint individual-group tracking
L. Bazzani, M. Cristani, V. Murino
In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012
FM dataset / video / bibtex
@inproceedings{Bazzani:CVPR12,
  title = {Decentralized Particle Filter for Joint
           Individual-Group Tracking},
  author = {Bazzani, L. and Cristani, M. and Murino, V.},
  booktitle = {IEEE Conference on Computer Vision and Pattern
           Recognition (CVPR)},
  year = {2012},
  month = {June},
  issn = {1063-6919}
}
    
This work presents a model to jointly track individuals and groups over time.
A very preliminary study was published at ICIP 2010
Re-identification with RGB-D sensors
B. I. Barbosa, M. Cristani, A. Del Bue, L. Bazzani, V. Murino
In 1st International Workshop on Re-Identification, 2012
RGBD-ID dataset / bibtex
@incollection{Barbosa:wECCV2012,
  title={Re-identification with RGB-D Sensors},
  author={Barbosa, I. B. and Cristani, M. and Del Bue, A. and
          Bazzani, L. and Murino, V.},
  year={2012},
  isbn={978-3-642-33862-5},
  booktitle={Computer Vision - ECCV 2012. Workshops and
          Demonstrations},
  volume={7583},
  series={Lecture Notes in Computer Science},
  editor={Fusiello, Andrea and Murino, Vittorio and Cucchiara, Rita},
  doi={10.1007/978-3-642-33863-2_43},
  url={http://dx.doi.org/10.1007/978-3-642-33863-2_43},
  publisher={Springer Berlin Heidelberg},
  pages={433-442},
  language={English}
}
   
We proposed a person re-identification descriptor exploting the advantages of depth sensors
Online bayesian non-parametrics for social group detection
M. Zanotto, L. Bazzani, M. Cristani, V. Murino.
In British Machine Vision Conference (BMVC), 2012
bibtex
@inproceedings{Zanotto:BVMC12,
  title = {Online Bayesian Non-parametrics for Social Group
           Detection},
  author = {Zanotto, M. and Bazzani, L. and Cristani, M. and
           Murino, V.},
  booktitle = {British Machine Vision Conference (BMVC)},
  year = {2012}
}
    
Learning attentional policies for object tracking and recognition in video with deep networks
L. Bazzani, N. de Freitas, H. Larochelle, V. Murino, J-A Ting
The 30th International Conference on Machine Learning (ICML), 2011
RBMtrack code / presentation / recorded talk / video / bibtex
@inproceedings{Bazzani:ICML11,
  title = {Learning attentional policies for object tracking and
           recognition in video with deep networks},
  author = {Bazzani, L. and de Freitas, N. and Larochelle, H. and
           Murino, V. and Ting, J-A},
  booktitle = {Proceedings of the 28th International Conference on
           Machine Learning (ICML-11)},
  year = {2011},
  address = {New York, NY, USA},
  editor = {Lise Getoor and Tobias Scheffer},
  month = {June},
  pages = {937--944},
  publisher = {ACM},
  series = {ICML '11}
}
    
The preliminary results of this work were presented at the Deep Learning and Unsupervised Feature Learning Workshop at NIPS 2010
Custom pictorial structures for re-identification
D. S. Cheng, M. Cristani, M. Stoppa, L. Bazzani, V. Murino
In British Machine Vision Conference (BMVC), 2011
CPS code / CAVIAR4REID dataset / video / bibtex
@inproceedings{Cheng:BMVC11,
  title = {Custom Pictorial Structures for Re-identification},
  author = {Cheng, D. S. and Cristani, M. and Stoppa, M. and
        Bazzani, L. and Murino, V.},
  booktitle = {British Machine Vision Conference (BMVC)},
  year = {2011}
}
    
Extension of SDALF (our CVPR 2010) through a finer subdivision of the human appearance in parts
Social interaction discovery by statistical analysis of F-formations
M. Cristani, L. Bazzani, G. Pagetti, A. Fossati, D. Tosato, A. Del Bue, G. Menegaz, V. Murino
In British Machine Vision Conference (BMVC), 2011
Project page / CoffeBreak dataset / bibtex
@inproceedings{Cristani:BMVC11,
  title = {Social interaction discovery by statistical
          analysis of F-formations},
  author = {Cristani, M. and Bazzani, L. and Pagetti, G. and
          Fossati, A. and Tosato, D. and Del Bue, A. and
          Menegaz, G. and Murino, V.},
  booktitle = {British Machine Vision Conference (BMVC)},
  year = {2011}
}
    
Towards computational proxemics: Inferring social relations from interpersonal distances
M. Cristani, G. Pagetti, A. Vinciarelli, L. Bazzani, G. Menegaz, V. Murino
In International Conference on Social Computing (SocialCom), 2011
bibtex
@inproceedings{Cristani:SCOM11,
  title = {Towards Computational Proxemics: Inferring Social
          Relations from Interpersonal Distances},
  author = {Cristani, M. and Pagetti, G. and Vinciarelli, A. and
         Bazzani, L. and Menegaz, G. and Murino, V.},
  booktitle = {International Conference on Social Computing
         (SocialCom)},
  year = {2011}
}
    
Multiple-shot person re-identification by hpe signature
L. Bazzani, M. Cristani, A. Perina, M. Farenzena, V. Murino
In International Conference on Pattern Recognition (ICPR), 2010
bibtex
@inproceedings{Bazzani:ICPR10,
  title = {Multiple-Shot Person Re-identification by HPE Signature},
  author = {Bazzani, L. and Cristani, M. and Perina, A. and Farenzena, M. and Murino, V.},
  booktitle = {20th International Conference on Pattern Recognition (ICPR)},
  year = {2010},
  month = {August},
  pages = {1413 -1416},
  bdsk-url-1 = {http://dx.doi.org/10.1109/ICPR.2010.349},
  doi = {10.1109/ICPR.2010.349},
  issn = {1051-4651}
}
    
IBM Best Student Paper Award track: Computer Vision.
For this reason, it was an invited article in the PRL journal listed above.
Person re-identification by symmetry-driven accumulation of local features
M. Farenzena, L. Bazzani, A. Perina, M. Cristani, V. Murino
In Conference on Computer Vision and Pattern Recognition (CVPR), 2010
SDALF code / video / bibtex
@inproceedings{Farenzena:CVPR10,
  title = {Person re-identification by symmetry-driven
         accumulation of local features},
  author = {Farenzena, M. and Bazzani, L. and Perina, A. and
         Murino, V. and Cristani, M.},
  booktitle = {IEEE Conference on Computer Vision and Pattern
        Recognition (CVPR)},
  year = {2010},
  month = {June},
  pages = {2360 -2367},
  bdsk-url-1 = {http://dx.doi.org/10.1109/CVPR.2010.5539926},
  doi = {10.1109/CVPR.2010.5539926},
  issn = {1063-6919}
}
    
This work has became one of the most popular person re-identification method to date and therefore standard baseline to compare with
Collaborative particle filters for group tracking
L. Bazzani, M. Cristani, V. Murino
In International Conference on Image Processing (ICIP), 2010
bibtex
@inproceedings{Bazzani:ICIP10,
  title = {Collaborative particle filters for group tracking},
  author = {Bazzani, L. and Cristani, M. and Murino, V.},
  booktitle = {17th IEEE International Conference on
         Image Processing (ICIP)},
  year = {2010},
  month = {September},
  pages = {837 -840},
  bdsk-url-1 = {http://dx.doi.org/10.1109/ICIP.2010.5653463},
  doi = {10.1109/ICIP.2010.5653463},
  issn = {1522-4880}
}
    
*Authors contributed equally

Book chapters

SDALF: modeling human appearance with symmetry-driven accumulation of local features
L. Bazzani, M. Cristani, V. Murino.
Person Re-identification, 2014.
SDALF code / bibtex
@incollection{Bazzani:REID14,
  title = {SDALF: Modeling Human Appearance with Symmetry-Driven
           Accumulation of Local Features},
  author = {Bazzani, L. and Cristani, M. and Murino, V.},
  booktitle = {Person Re-identification},
  year = {2014}
}
    
This work subsumes our CVPR 2010 and CVIU 2013 papers.
Analyzing groups: a social signaling perspective
L. Bazzani, M. Cristani, G. Paggetti, D. Tosato, G. Menegaz, and V. Murino.
Video Analytics for Business Intelligence, 2012.
IRPM code / bibtex
@incollection{Bazzani:VABI12,
  title = {Analyzing Groups: A Social Signaling Perspective},
  author = {Bazzani, L. and Cristani, M. and Paggetti, G. and
            Tosato, D. and Menegaz, G. and Murino, V.},
  booktitle = {Video Analytics for Business Intelligence},
  year = {2012},
  pages = {271-305},
  volume = {409}
}
    
This work subsumes our ICIP 2010 and ICIAP 2009 papers

Dissertations

Beyond Multi-target tracking: statistical pattern analysis of people and groups
L. Bazzani
Ph.D. dissertation, The University of Verona, 2012
bibtex
@phdthesis{Bazzani:PhD12,
  title = {Beyond Multi-target tracking: statistical pattern
           analysis of people and groups},
  author = {Bazzani, L.},
  school = {University of Verona},
  year = {2012}
}
    
Reviewers: A. Del Bimbo and R. T. Collins

Particle filtering approaches for multi-target tracking in video surveillance applications
L. Bazzani
M.S. thesis, The University of Verona, 2008
bibtex
@unpublished{Bazzani:Mscthesis,
  title = {Particle Filtering Approaches for Multi-target
           Tracking in Video Surveillance Applications},
  author = {Bazzani, L.},
  year = {2008}
}
    
Work presented at ICIP 2009 and PETS 2009

Techniques for the analysis and the classification of MRI for searching pathology with application to mental health
L. Bazzani
B.S. thesis, The University of Verona, 2006
bibtex
@unpublished{Bazzani:Bscthesis,
  title = {Techniques for the Analysis and the Classification
           of MRI for Searching Pathology with Application to
           Mental Health},
  author = {Bazzani, L.},
  year = {2006}
}
    
Work presented at the Joint Annual Meeting ISMRM-ESMRMB 2007

I like this website and this