Institutional Repository of Vilnius University
Objektų aptikimas naudojant nedidelėms mokymo duomenų aibėms pritaikytus giliuosius neuroninius tinklus /
Abstract
dc:descriptionThe object detection task, tackled using neural networks, can be adapted to detect distracted drivers. However, this task requires the use of a dataset that portrays distracted drivers. Since such a database is not publicly accessible, a new dataset of distracted drivers pictured from the front was created. During this research experiment, four different few-shot object detection neural network architectures were trained. One of these architectures, MPSR, trained on a dataset adapted to the PASCAL VOC database, using a 10-shot training configuration, manages to achieve a 62.1% average precision. This neural network sometimes assigns several bounding boxes, each belonging to different classes, to one object. In that case, by deleting all of the bounding boxes, except for the most likely one, the results generated by network are accurate enough to have practical usage.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Globis, Einartas Jonas,
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:ELABAETD146241266&prefLang=en_US