Název:	From Sources to Solutions: Enhancing Object Detection Models through Synthetic Data
Autoři:	Bartolovic, Eduard Höfer, Tobias Hage, Clemens Nischwitz, Alfred
Citace zdrojového dokumentu:	WSCG 2024: full papers proceedings: 32. International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, p. 263-272.
Datum vydání:	2024
Nakladatel:	Václav Skala - UNION Agency
Typ dokumentu:	konferenční příspěvek conferenceObject
URI:	http://hdl.handle.net/11025/57398
ISSN:	2464–4625 (online) 2464–4617 (print)
Klíčová slova:	syntetické generování dat;augmentace dat;randomizace domény;detekce objektu;YOLOv5;SegmentAnything
Klíčová slova v dalším jazyce:	synthetic data generation;data augmentation;domain randomization;object detection;YOLOv5;SegmentAnything
Abstrakt v dalším jazyce:	Object detection, a fundamental task in computer vision, plays a crucial role in various applications such as au tonomous driving, surveillance, and robotics. However, training models for this task require vast amounts of high-quality data, often involving labor-intensive manual labeling. Synthetic data, a promising alternative, re mains an active area of research. This paper presents a comprehensive exploration of different object sources for the use of synthetic data in enhancing object detection models. We investigate various synthetic data gen eration techniques to implant objects into a scene, with a focus on enhancing training data diversity. These ob jects are either gathered from the training dataset itself using SegmentAnything as a new supervised self aug mentation technique or imported from external sources, including a photobox with a rotating table and web scraping of online shops. Moreover, our study delves into the development of a placement logic that gradu ally evolves from placing objects randomly to placing objects in physically correct orientations to mimic the real world data. We investigate the use of different blending techniques. The outcome of our study demon strates that synthetic images, when integrated with an existing real training set, substantially improve the ob ject recognition accuracy of the model without compromising inference time. Our code can be found at https://github.com/EduardBartolovic/synthetic-data-generation
Práva:	© Václav Skala - UNION Agency
Vyskytuje se v kolekcích:	WSCG 2024: Full Papers Proceedings

Všechny záznamy v DSpace jsou chráněny autorskými právy, všechna práva vyhrazena.