Zusammenfassung
Diese Disse tation beschäftigt sich mit der bildbasierten Kopfposenschätzung mittels Deep Learning und Lernen aus synthetischen Daten. Synthetische Daten bieten viele Vorteile gegenüber realen Daten, die aufwendig in der realen Welt gesammelt werden müssen. Allerdings erreicht Deep Learning mit synthetischen Daten oft nicht die Genauigkeit, die mit realen Daten erreicht werden könnte. Um dieses Problem zu lösen, werden in dieser Arbeit zwei Methoden zur unüberwachten Domänenadaption vorgestellt. Diese verbessern die Genauigkeit, die durch Training mit synthetischen Daten auf realen Daten erreicht werden kann, erheblich. Darüber hinaus wird in dieser Arbeit der Einfluss verschiedener Faktoren auf die Bewertung von Schätzalgorithmen für Kopfposen analysiert. Contents 1 introduction 1 1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Challenges in Domain Adaptation with Synthetic Data and Continuous Label Spaces . . . . . . . . . . . . . . . . . 4 1.3 Objectives and Contributions . . . . . . . . . . . . . . . . . 7 1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2 fundamentals 11 2.1 Machine Learning with Deep Neural Netw...
Schlagworte
Kopfposenschätzung Domänenadaption synthetische Daten Deep Learning Regression head pose estimation domain adaptation synthetic data deep learning regression- Kapitel Ausklappen | EinklappenSeiten
- 1–10 1 Introduction 1–10
- 11–30 2 Fundamentals 11–30
- 61–98 5 Evaluation 61–98
- 99–102 6 Conclusion 99–102
- 103–103 a Appendix 103–103
- 103–108 a.1 Additional Tables 103–108
- 109–110 a.2 Additional Figures 109–110
- 111–128 Bibliography 111–128
8 Treffer gefunden
- „... of partial domain adaptation (PDA) for regression.This chapter shows how we transform adversarial ...” „... following subsections.3.2.1 DANN for regressionWe start by reintroducing the DANN objective but adapt it by ...” „... consequence, itseems to be reasonable to apply them to regression tasks as well. Regres-sion tasks have a ...”
- „... (HPE) is a regression task that ideally fits thepresented scenario. Taking pictures of faces is easy ...” „... DA. Regression tasks usually implydifferences in label distributions between domains that cause ...” „... not truefor regression tasks, which have continuous label spaces. In a continuouslabel space, the ...”
- „... regression tasks including HPE, is still an openchallenge for DA. Unfortunately, related work does not ...” „... contrast tomost DA research, we focus on regression tasks and, in doing so, arefilling an important ...” „... geodesic loss, and 6D pose rep-resentation seem to reliably improve pose regression. As pose ...”
- „... ]. At thispoint, regression is rarely investigated in semi-supervised learning. It istherefore exciting ...” „... to explore whether these methods can also be used fordomain adaptation with a regression task. On top ...” „... regression task. The objective is to find a smalldistance between predicted poses g(x′) and ground truth ...”
- „... Dokumentation: Kopfposenschätzung – Domänenadaption – synthetische Daten – Deep Learning – RegressionKeywords ...” „... : head pose estimation – domain adaptation – synthetic data – deep learning – regressionDiese ...” „... Adversarial Domain Adaptation for Continuous Label Spaces 363.2.1 DANN for regression ...”
- „... : classi-fication and regression. For classification the algorithm has to accuratelyassign data into specific ...” „... simultaneously. For regression the algorithmhas to estimate the relationships between input variables (often ...” „... predictive function that helps us solve a problemlike regression or classification [49].Predictive Function ...”
- „... . Hoffmann, S. Tripathi, andM. J. Black. “AGORA: Avatars in Geography Optimized for Re-gression Analysis.” In ...” „... , and R. Horaud.“Deep mixture of linear inverse regressions applied to head-poseestimation.” In ...” „... . Lathuilière, P. Mesejo, X. Alameda-Pineda, and R. Horaud. “AComprehensive Analysis of Deep Regression.” In ...”
- „... ], adapted byus for regression in Section 3.2.1, and PADACO. This allows us to comparethe performance of ...” „... Section 3.2.2 and Section 3.2.1to work for regression tasks. PADACO, DANN, and PADA-like share ...” „... the regressionnetwork r is the replaced last layer of ResNet18. The domain discrim-inator d is a fully ...”