Home page of Alessandro Perina

Contact

alessandro.perina [at] gmail [dot] com

skype: ale.perina

Education

Dec. 2004 - BSc in IT, University of Verona, Verona

Sept. 2006 - MSc in Intelligent systems, University of Verona, Verona

Nov. 2010 - Ph.D. in Computer Science, University of Verona, Verona - tutor: Prof. Vittorio Murino. Thesis title: "Generative models for classification "

Positions

Sept. 2006 - Dec. 2006, Research Assistant, Vision Algorithms for industrial applications, VIPS Lab, Verona

Jan. 2007 - Dec. 2009 Ph.D. Student (Scholarship funded by D.M. 198/03, Ministero della Ricerca, Repubblica Italiana). Graduate School in Sciences, Engeenering and Medicine, University of Verona.

Feb. 2008: - Apr. 2008, Research Intern, Microsoft Research, Redmond, WA. Machine Learning and Applied Statistics Group

Jun. 2009 - Sept. 2009, Research Intern, Microsoft Research, Redmond, WA. eScience Group

Jan. 2010 - Jun. 2010, Research Associate, SIMBAD Project, University of Verona, Verona

Jun. 2010 - Now, Research Scholar, Microsoft Research, Redmond, WA - Mentor: Dr. Nebojsa Jojic, eScience Group


My main goal is to solve computer vision problems at multiple levels by using and developing machine learning techniques, taking inspiration from the theories of human learning. I am currently working on summarizing one's visual input using epitomic or epitomic-like representations. Below a video of one of the first satisfactory results...


Projects: project pages: MATLAB code, datasets used, ground-truth. See videos of my works [here]

Snapshot
The Counting Grid

Counting Grid, FFT-Version used in the paper [code]

Code is a beta version even well written. Please report any problem and I will send you a newer version of the code. Several versions of the code will follow as several extensions of the work are in submission to major conferences.

Probabilistic index maps and STEL models

Stel Component Analysis (PIM) code [code]

Larry Video used for CVPR09 paper (frames + GroundTruth) [download]

Code is a beta version and it is NOT (time) optimzed. It requires Tom Minka's lightspeed and fastfit toolboxes. Send me an email if you have questions (expecially on the training regime I've used). I mean my code as starting point for you own implementation.

Free Energy Score Space (FESS)

Click on the image for the project page [code + additional information]

Free eneregy score space have been recently improved by Xion Li. Check out his CVPR 2011 paper. "Hybrid Generative-Discriminative Classification using Posterior Divergence"!

SenseCam Dataset

I am currently labeling 10% of the dataset. Labels will be available in the next few months.

Other projects

  1. Pedestrian Re-identification (ICPR 2010, CVPR 2010): Loris webpage
  2. Geo-Located images (CVPR 2008): Marco webpage
  3. Topic models for bioinformatics (SAC 2010, PRIB 2010): Manuele webpage @IIT