Vehicle detection on small scale data by generative data augmentation

Kumdakcı, Hilmi
Scarcity of training data is one of the prominent problems for deep neural networks,which commonly require high amounts of data to display their potential. Data aug-mentation techniques are frequently applied during the pre-training and training phasesof deep neural networks to overcome the problem of having insufficient data for train-ing. These techniques aim to increase a neural network’s generalization performanceon unseen data by increasing the number of training samples and provide a more rep-resentative distribution to the system during training. In this work, we focus on im-proving vehicle detection in aerial images by proposing a data augmentation methodthat does not need any extra supervision than the bounding box annotations of thevehicle instances in the training data. The methods we used are based on a condi-tional Generative Adversarial Network (cGAN). The proposed method is not exclu-sive and can be used in association with classical augmentation techniques to furtherimprove object detection performance. We showed that the proposed data augmenta-tion method increases the Average Precision by up to 25.2%, 32.7%, and 25.7% whenintegrated with Pluralistic, PSGAN, and DeepFill respectively.


Generative Data Augmentation for Vehicle Detection in Aerial Images
Kumdakçı, Hilmi; Öngün, Cihan; Temizel, Alptekin (null; 2021-01-15)
Scarcity of training data is one of the prominent problemsfor deep networks which require large amounts data. Data augmentationis a widely used method to increase the number of training samples andtheir variations. In this paper, we focus on improving vehicle detectionperformance in aerial images and propose a generative augmentationmethod which does not need any extra supervision than the boundingbox annotations of the vehicle objects in the training dataset. The pro-posed method increases the perf...
Generation and modification of 3D models with deep neural networks
Öngün, Cihan; Temizel, Alptekin; Department of Information Systems (2021-9)
Artificial intelligence (AI) and particularly deep neural networks (DNN) have become very hot topics in the recent years and they have been shown to be successful in problems such as detection, recognition and segmentation. More recently DNNs have started to be popular in data generation problems by the invention of Generative Adversarial Networks (GAN). Using GANs, various types of data such as audio, image or 3D models could be generated. In this thesis, we aim to propose a system that creates artificial...
Spatio-temporal pattern and trend extraction on Turkish meteorological data
Goler, Isil; Karagöz, Pınar; Yazıcı, Adnan (2012-12-01)
Due to increasing amount of spatio-temporal data collected from various applications, spatio-temporal data mining has become a demanding and challenging research field requiring development of novel algorithms and techniques for successful analysis of large spatio-temporal databases. In this study, we propose a spatio-temporal mining technique and apply it on meteorological data, which has been collected from various weather stations in Turkey. In addition, we introduce one more mining level on the extracte...
New Techniques in Profiling Big Datasets for Machine Learning with a Concise Review of Android Mobile Malware Datasets
CANBEK, Gurol; SAĞIROĞLU, ŞEREF; Taşkaya Temizel, Tuğba (2018-12-04)
As the volume, variety, velocity aspects of big data are increasing, the other aspects such as veracity, value, variability, and venue could not be interpreted easily by data owners or researchers. The aspects are also unclear if the data is to be used in machine learning studies such as classification or clustering. This study proposes four techniques with fourteen criteria to systematically profile the datasets collected from different resources to distinguish from one another and see their strong and wea...
Domain adaptation on graphs by learning graph topologies: theoretical analysis and an algorithm
Vural, Elif (The Scientific and Technological Research Council of Turkey, 2019-01-01)
Traditional machine learning algorithms assume that the training and test data have the same distribution, while this assumption does not necessarily hold in real applications. Domain adaptation methods take into account the deviations in data distribution. In this work, we study the problem of domain adaptation on graphs. We consider a source graph and a target graph constructed with samples drawn from data manifolds. We study the problem of estimating the unknown class labels on the target graph using the...
Citation Formats
H. Kumdakcı, “Vehicle detection on small scale data by generative data augmentation,” M.S. - Master of Science, Middle East Technical University, 2021.