See veebileht kasutab küpsiseid kasutaja sessiooni andmete hoidmiseks. Veebilehe kasutamisega nõustute ETISe kasutustingimustega. Loe rohkem
Olen nõus
"Muu" projekt VLTTI16119
VLTTI16119 "A.G.E." (11.04.2016−15.07.2016); Vastutav täitja: Gholamreza Anbarjafari; Tartu Ülikool, Loodus- ja täppisteaduste valdkond, tehnoloogiainstituut (partner); Finantseerija: Super Appli, Inc.; Eraldatud summa: 17 442 EUR.
Age and Gender estimation
Teadus- ja arendusprojekt
välis TA ettevõtlusleping
ETIS klassifikaatorAlamvaldkondCERCS klassifikaatorFrascati Manual’i klassifikaatorProtsent
4. Loodusteadused ja tehnika4.6. ArvutiteadusedT121 Signaalitöötlus 1.1. Matemaatika ja arvutiteadus (matemaatika ja teised sellega seotud teadused: arvutiteadus ja sellega seotud teadused (ainult tarkvaraarendus, riistvara arendus kuulub tehnikavaldkonda)100,0
Super Appli, Inc.Jaapan
11.04.2016−15.07.201617 442,00 EUR
17 442,00 EUR

We are going to explain how we will implement age estimation systems step by step. Step 1 : Database Collecting First of all, we will collect about half million Chinese Face photos, and their apparent age labels. This process is most important one for accuracy of the system. Step 2 : Caffe Deep Learning Framework We are going to develop our age estimation system based on deep learning. Deep learning is currently one of the best machine learning method. Nowadays computers¿ power are getting dramatically improved aid of GPU. Although Deep learning is old method, it waited until the computer became more powerful. Caffe Framework is most famous and powerful deep learning method. So we will compile and have some changes on framework for our system requirements. Step 3 : Feature Extraction We will write the software which will take all faces and resize them to correct parameters. Then we will use VGG-16 (Very Deep Convolutional Networks for Large-Scale Visual Recognition) model to get features of image. Step 4 : Training After feature extraction, we will use our labels and features to create caffe type database. Adjusted Caffe framework will train all system with using GPU. Step 5 : Prediction In order to predict the age, we will adopt softmax expected value algorithm. We will define 0 to 100 ages. In every age value, there will be probability of estimated age. Thus we will be able to use below formula: E(O) = yi * oi yi is the estimated age of every age label, and oi is the value of age label. Step 6 : Auto Training We will keep collecting photos in order to increase performance of our system. We will write corn job program in Linux. After late time when the number of users will be low, our system will start to train itself. In that way, every single day our system will have better accuracy.