magistrsko delo
Abstract
Magistrsko delo se ukvarja z ocenjevanjem starosti osebe na osnovi digitalnih posnetkov z uporabo konvolucijskih nevronskih mrež. Razvit in implementiran je bil lasten model konvolucijske nevronske mreže za ocenjevanje starosti osebe iz digitalnega posnetka. Kot osnova za naš model je bila uporabljena in modificirana obstoječa arhitektura konvolucijske nevronske mreže VGG-Face, namenjena razpoznavanju obrazov. Za učenje in testiranje sta bili uporabljeni bazi podatkov IMDB-WIKI in FG-NET. Na bazi podatkov IMDB-WIKI je bila dosežena povprečna napaka med dejansko in ocenjeno starostjo 6,7 leta, na bazi podatkov FG-NET pa z validacijsko metodo »izpusti-eno-osebo« izračunana povprečna napaka med dejansko in ocenjeno starostjo 3,9 leta. Dobljeni rezultati so primerljivi oziroma le malo zaostajajo za najuspešnejšimi metodami za ocenjevanje starosti osebe z digitalnega posnetka. Na tej osnovi se naš model ocenjuje kot primeren za uporabo v produkcijskih rešitvah.
Keywords
računalniški vid;konvolucijske nevronske mreže;globoko učenje;ocenjevanje starosti;magistrske naloge;
Data
Language: |
Slovenian |
Year of publishing: |
2021 |
Typology: |
2.09 - Master's Thesis |
Organization: |
UM FERI - Faculty of Electrical Engineering and Computer Science |
Publisher: |
[T. Krel] |
UDC: |
004.85:004.932(043.2) |
COBISS: |
54782979
|
Views: |
599 |
Downloads: |
60 |
Average score: |
0 (0 votes) |
Metadata: |
|
Other data
Secondary language: |
English |
Secondary title: |
Person age estimation based on digital images using convolutional neural networks |
Secondary abstract: |
In the master’s thesis, we focused on person age estimation based on digital images using convolutional neural networks. We developed and implemented our own convolutional neural network model, used to estimate age of a person from a digital image. As a base for our model, we used and modified the existing convolutional neural network architecture VGG-Face, used for face recognition. For learning and testing, the IMDB-WIKI and FG-NET datasets were used. With the IMDB-WIKI dataset, we can achieve the average error between the actual and the estimated age of 6.7 years, while using the dataset FG-NET, we can calculate the average error between the actual and the estimated age of 3.9 years, employing the »leave-one-person-out« validation method. The obtained results are comparable to or only slightly behind the most successful methods for age estimation from a digital image. On this basis, we evaluate our model as suitable for use in production solutions. |
Secondary keywords: |
computer vision;convolutional neural networks;deep learning;age estimation; |
Type (COBISS): |
Master's thesis/paper |
Thesis comment: |
Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije |
Pages: |
VII, 44 f. |
ID: |
12415620 |