Object Detection Using Synthetic Training Data
(2020) In Master’s Theses in Mathematical Sciences FMAM05 20201Mathematics (Faculty of Engineering)
- Abstract
- Annotating data for machine learning purposes can be very inefficient and time-consuming. In this paper we introduce a pipeline for generating a 3d scene from a simple image. We discuss and develop the first step which includes a 3d network being able to recognise any particular object. We focus on how synthetic data can be used to make the annotating process simpler. We use generated synthetic images and train two different networks (YOLO and DOPE) and study their performances in order to learn how to create better training data. Finally, we conclude that domain randomization is very useful for attaining good results and discuss the diculties when studying training data.
- Popular Abstract (Swedish)
- Två olika artificiella intelligenser har tränats för att kunna känna igen leksaker i bilder och film. För att undvika processen att annotera tusentals bilder används ett program som kan generera annoterade bilder med 3D modeller av leksakerna i realistiska och orealistiska datorgjorda miljöer. Vi testar vilka miljöer som ger oss bäst resultat och kommer fram till att en kombination av realistiska och orealistiska bilder på leksakerna ger bäst resultat.
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9010050
- author
- Pettersson, Assar LU and Stjernström, Filip LU
- supervisor
-
- Karl Åström LU
- David Gillsjö LU
- organization
- course
- FMAM05 20201
- year
- 2020
- type
- H2 - Master's Degree (Two Years)
- subject
- publication/series
- Master’s Theses in Mathematical Sciences
- report number
- LUTFMA-3400-2020
- ISSN
- 1404-6342
- other publication id
- 2020:E13
- language
- English
- id
- 9010050
- date added to LUP
- 2020-06-24 14:36:50
- date last changed
- 2020-06-24 14:36:50
@misc{9010050, abstract = {{Annotating data for machine learning purposes can be very inefficient and time-consuming. In this paper we introduce a pipeline for generating a 3d scene from a simple image. We discuss and develop the first step which includes a 3d network being able to recognise any particular object. We focus on how synthetic data can be used to make the annotating process simpler. We use generated synthetic images and train two different networks (YOLO and DOPE) and study their performances in order to learn how to create better training data. Finally, we conclude that domain randomization is very useful for attaining good results and discuss the diculties when studying training data.}}, author = {{Pettersson, Assar and Stjernström, Filip}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master’s Theses in Mathematical Sciences}}, title = {{Object Detection Using Synthetic Training Data}}, year = {{2020}}, }