Transfer Learning Methods for Training Person Detector in Drone Imagery
| Autor: | Marina Ivašić-Kos, Sasa Sambolek |
|---|---|
| Přispěvatelé: | Arai, Kohei |
| Jazyk: | angličtina |
| Rok vydání: | 2022 |
| Předmět: |
Computer science
business.industry Transfer learning YOLO v4 Person detection Drone dataset Detector Training (meteorology) Machine learning computer.software_genre Drone Task (project management) Data set Generalization (learning) Artificial intelligence Transfer of learning business computer Search and rescue |
| Zdroj: | Lecture Notes in Networks and Systems ISBN: 9783030821951 IntelliSys (2) |
| Popis: | Deep neural networks achieve excellent results on various computer vision tasks, but learning models require large amounts of tagged images and often unavailable data. An alternative solution of using a large amount of data to achieve better results and greater generalization of the model is to use previously learned models and adapt them to the task at hand, known as transfer learning. The aim of this paper is to improve the results of detecting people in search and rescue scenes using YOLOv4 detectors. Since the original SARD data set for training human detectors in search and rescue scenes are modest, different transfer learning approaches are analyzed. Additionally, the VisDrone data set containing drone images in urban areas is used to increase training data in order to improve person detection results. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|