Wednesday, 21 Feb 2018
You are here:

Subscribe to our Newsletter

How many eyes has a typical person? (ex: 1)

Twitter Feed

EMail Print


Jean Luc Dugelay & Natacha Ruchaud (Eurecom) visiting Prof. Touradj Ebrahimi (EPFL)
04.12.2014 - 06.12.2014
Internal Visits


During this visit, Partners discussed and prepared the plan for joint collaboration and planned joint publication. The main idea behind the work is to compare human vision with computer for gender recognition from more or less filtered body. Original plan included working on drone videos but finally another database, PETA which contains 10 others database, has been identified. Convolutional neural network has been used to learn a model and then to test. Crowdsourcing on microworker will be done to fill our survey.
First both Partners had to sort PETA dataset because some data was misannoted. Then test images has been selected for the crowdsourcing task. In effect, the test dataset must be balanced between female and male. CNN is learned on PETA dataset and tested on selected original images. Deidentifications are applied with several filters on selected test images. The last step is the classification by the computer to predict gender with CNN from this filtering images.
Last part of the job is to evaluate gender classification by human vision. To do it the crowdsourcing via a website: will be applied.
The result of the visit was the plan involving EPFL focusing on Crowdsourcing and Eurecom sorting PETA dataset, selecting test data, learning and testing with CNN and applying filters. The visit has contributed to WP2 and WP6 activities.