Compact hash codes and data structures for efficient mobile visual search

We present an efficient method for mobile visual search that exploits compact hash codes and data structures for visual features retrieval. The method has been tested on a large scale standard dataset of one million SIFT features, showing a retrieval performance comparable or superior to state-of-the-art methods, and a very high efficiency in terms of memory consumption and computational requirements. These characteristics make it suitable for application to mobile visual search, where devices have limited computational and memory capabilities.

Article : Compact hash codes and data structures for efficient mobile visual search

Reddit Crawler in Python

Code of a Reddit Crawler written in Python


Reddit /ˈrɛdɪt/,[4] stylized as reddit,[5] is an entertainment, social networking service and news website where registered community members can submit content, such as text posts or direct links. Only registered users can then vote submissions “up” or “down” to organize the posts and determine their position on the site’s pages. Content entries are organized by areas of interest called “subreddits”.

Reddit was founded by University of Virginia roommates Steve Huffman and Alexis OhanianCondé Nast Publications acquired the site in October 2006. Reddit became a direct subsidiary of Condé Nast’s parent company, Advance Publications, in September 2011. As of August 2012, Reddit operates as an independent entity, although Advance is still its largest shareholder.[6] Reddit is based in San Francisco, California. In October 2014 Reddit received $50 million in funding from Snoop Lion and other investors.[7]



link to the CrawlerReddit

C++ Ransac Method Implementation

Ransac Method implemented in C++ for robust keypoints estimation in Computer Vision


Random sample consensus (RANSAC) is an iterative method to estimate parameters of a mathematical model from a set of observed data which contains outliers. It is a non-deterministic algorithm in the sense that it produces a reasonable result only with a certain probability, with this probability increasing as more iterations are allowed. The algorithm was first published by Fischler and Bolles at SRI International in 1981.

A basic assumption is that the data consists of “inliers”, i.e., data whose distribution can be explained by some set of model parameters, though may be subject to noise, and “outliers” which are data that do not fit the model. The outliers can come, e.g., from extreme values of the noise or from erroneous measurements or incorrect hypotheses about the interpretation of data. RANSAC also assumes that, given a (usually small) set of inliers, there exists a procedure which can estimate the parameters of a model that optimally explains or fits this data.



Here the code (under GitHub-Gist)  RansacMethodCPP

Vision and Multimedia Reading Group: DeCAF

I presented an interesting paper during the Vision and Multimedia Reading Group about DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition (pdf).

It is a complete evaluation about  features extracted from the activation of a deep convolutional network trained with a large scale dataset.

This a work of Jeff Donahue, Yangqing Jia, Oriol Vinyals, Judy Hoffman, Ning Zhang, Eric Tzeng, Trevor Darrell from Berkeley University



A framework for itinerary personalization in cultural tourism of smart cities

Smart tourism in cities of art is a personalized user experience that exploits smart city infrastructures to offer increased opportunities of visit and services and time optimization. Traditionally, this capability requires the availability of personal mobile systems and geolocalization, augmented with some smart computing that provides the due information and functions at the right time and location of the visit. However effective smart tourism should also account for the fact that they exist different user requirements at different stages of the visit and that interests and requirements not only differ from one user to the other but also may change through time for each individual user. According to this, an effective framework for smart tourism should offer the possibility of an easy definition of individual user visits and offer to each user the capability of making changes or updates to his/her visit plan during the visit. It should also consider the possibility that different devices are offered and used at the different stages of the visit. In this work we present the prototype of a framework where different devices are used for the definition and modification of a personalized visit. In particular it exploits a wall mounted touchscreen in a visitor center which permits the early definition of a visit plan and a mobile device which allows online updates and changes of the planes well as display of geolocalized information during the time of the visit. An application server platform and a network infrastructure allows to record user activities as well as search and retrieve personalized data.




Article : A framework for Itinerary Personalization in Cultural Tourism of Smart Cities

Copyright © 2018  Simone Ercoli  All Rights Reserved.