Institutional Repository of Vilnius University
Dirbtinis neuroninis tinklas gylio kadre nustatymui /
Abstract
dc:descriptionArtificial neural networks for depth estimation often use an encoder-decoder architecture that utilizes already developed object classification models to solve the problem of depth estimation. This paper investigates which already developed object classification model is most suitable for use in the development of a depth estimating neural network. A review of the scientific literature for the depth estimating neural networks is performed. The models of object classification in image, which are reused to create these networks, as well as the data sets used for network training are also reviewed. Six modifications of the neural network of the encoder-decoder architecture were implemented using the classification models of ResNetV2 and DenseNet in the encoder part. The comparative analysis revealed that the most suitable model for depth estimation from all the performed modifications is DenseNet201.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Virbalaitė, Dovilė,
- Contributors dc:contributor
-
- Uosis, Karolis
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD81704728&prefLang=en_US