Training data is the bottleneck for training Convolutional Neural Networks. A larger dataset gives better accuracy though also needs longer training time. It is shown by finetuning neural networks on synthetic rendered images, that the mean average precision increases. This method was applied to two different datasets with five distinctive objects in each. The first dataset consisted of random objects with different geometric shapes. The second dataset contained objects used to assemble IKEA furniture. The neural network with the best performance, trained on 5400 images, achieved a mean average precision of 0.81 on a test which was a sample of a video sequence. Analysis of the impact of the factors dataset size, batch size, and numbers of epochs used in training and different network architectures were done. Using synthetic images to train CNN’s is a promising path to take for object detection where access to large amount of annotated image data is hard to come by.

<p>By focusing on the potential of the Wii Remote, we have implemented a broad spectrum of concept ideas into the same package, in an effort to give a good overview of the Wii Remote’s properties such as mobility, direct manipulation and generally high affordance. The purpose of this work was to find a concept on how a Wii Remote can be used as a tool for the industry, outside the domains of gaming and entertainment. The environment for our investigation was ABB’s Robot Studio which is a simulation tool for industrial robots. Creating a concept with today’s products enabled us to discuss a present solution and also a possible future in form of a redesign rationale that we exemplified with a set of scenarios.</p>

<p>Detta examensarbete behandlar ett experiment som Tom Smith utförde vid sitt magisterarbete, att utveckla en Kheperarobot som ska utföra en fotbollsuppgift av enklare modell. Dock koncentrerar sig detta arbete mer på en evaluering av artificiella neurala nätverk för detta problem. De olika typerna av ANN-arkitekturer som har använts till detta arbete är förutom Tom Smiths, baserade på arkitekturer från en artikel skriven av Stefano Nolfi.</p><p>De resultat som har uppnåtts visar att precis som i Stefano Nolfis artikel är det en arkitektur med "spontan modularitet" som visar sig fungera bäst av de arkitekturer som undersökts, även till detta problem.</p>

<p>Forskning angående styrsystem inom evolutionär robotik fokuserar ofta på vad som går att uträtta med självlärda styrsystem, men inte vad beteendebaserade styrsystem klarar av. I detta projekt utförs systematiska tester för att jämföra beteendebaserade och självlärda styrsystem inom evolutionär robotik. Benämningen beteendebaserade styrsystem används för styrsystem som består av flera underliggande moduler, där robotens övergrippande beteende är uppdelat i flera moduler och där respektive modul har ansvar för ett specifikt beteende eller funktion. Självlärt styrsystem referar i detta sammanhang till styrsystem som inte explicit består av moduler som har åstakommits eller bestämts i förväg.</p><p>Från resultaten framgår det att både självlärda och beteendebaserade styrsystem klarar av att lösa problemen de ställs inför. För beteendebaserade styrsystem krävs dock större ansvar från experimentatorn, vars inflytande ökar på grund av att det är flera moduler som skall skapas och koordineras. De beteendebaserade styrsystemen i projektet använder en beslutsenhet för att hantera när och vilken modul som ska aktiveras och tar bort detta ansvar från experimentatorn. Beslutsenheten ger styrsystemet en mer smidig övergång mellan de olika modulerna och hittar en lämplig användning av modulerna än utan en beslutsenhet. Från resultatet av projektet framgår det att fler systematiska tester angånde beteendebaserade och självlärda styrsystem behövs för att få en bättre förståelse över när och hur de olika styrsystemen bör användas.</p>

Rheinische Fachhochsch., Diplomarbeit--Köln, 2001.

Zugl.: Hannover, Universiẗat, Diss., 2008. / Mit e. engl. u. dt. Zsfassung.

Zugl.: Clemson, Clemson Univ., Diss., 2007 / Hergestellt on demand

Zugl.: Würzburg, Univ., Diss., 2009.

Thermal spectrum cameras are gaining interest in many applications due to their long wavelength which allows them to operate under low light and harsh weather conditions. One disadvantage of thermal cameras is their limited visual interpretability for humans, which limits the scope of their applications. In this thesis, we try to address this problem by investigating the possibility of transforming thermal infrared (TIR) images to perceptually realistic visible spectrum (VIS) images by using Convolutional Neural Networks (CNNs). Existing state-of-the-art colorization CNNs fail to provide the desired output as they were trained to map grayscale VIS images to color VIS images. Instead, we utilize an auto-encoder architecture to perform cross-spectral transformation between TIR and VIS images. This architecture was shown to quantitatively perform very well on the problem while producing perceptually realistic images. We show that the quantitative differences are insignificant when training this architecture using different color spaces, while there exist clear qualitative differences depending on the choice of color space. Finally, we found that a CNN trained from daytime examples generalizes well on tests from night time.

This master thesis explores the possibility of using generative Adversarial Networks (GANs) to refine labeled synthetic code images to resemble real code images while preserving label information. The GAN used in this thesis consists of a refiner and a discriminator. The discriminator tries to distinguish between real images and refined synthetic images. The refiner tries to fool the discriminator by producing refined synthetic images such that the discriminator classify them as real. By updating these two networks iteratively, the idea is that they will push each other to get better, resulting in refined synthetic images with real image characteristics. The aspiration, if the exploration of GANs turns out successful, is to be able to use refined synthetic images as training data in Semantic Segmentation (SS) tasks and thereby eliminate the laborious task of gathering and labeling real data. Starting off from a foundational GAN-model, different network architectures, hyperparameters and other design choices are explored to find the best performing GAN-model. As is widely acknowledged in the relevant literature, GANs can be difficult to train and the results in this thesis are varying and sometimes ambiguous. Based on the results from this study, the best performing models do however perform better in SS tasks than the unrefined synthetic set they are based on and benchmarked against, with regards to Intersection over Union.