MaActionGen

Chair for Computer Aided Medical Procedures & Augmented Reality
Lehrstuhl für Informatikanwendungen in der Medizin & Augmented Reality

THIS WEBPAGE IS DEPRECATED - please visit our new website

Action Recognition and Generation With RNNs

Human action understanding and generation is a high-level concept and very hot topic in computer vision. Deep learning methods made significant research progresses in this area. In this task, we aim to understand and generate realistic human actions from a given video input. We will use adversarial domain adaptation algorithm [2] to train Recurrent Neural Networks [1]. More specifically, Generative Adversarial Networks(GANs) will be used to reduce the discrepancy between training the network and sampling from it over multiple time steps. We will compare the accuracy and efficiency of this approach with other recent object detection methods.



Supervision: Huseyin Coskun, Christian Rupprecht, Federico Tombari


For further information please contact: Huseyin Coskun

Literature :
[1] Hochreiter, Sepp, and Jürgen Schmidhuber. "Long short-term memory." Neural computation 9.8 (1997): 1735-1780.
[2] Goodfellow, Ian, et al. "Generative adversarial nets." Advances in neural information processing systems. 2014.
[3] Lamb, Alex M., et al. "Professor forcing: A new algorithm for training recurrent networks."Advances In Neural Information Processing Systems. 2016.


Image source:http://pointclouds.org/gsoc/

ProjectForm
Title: Action Recognition and Generation With RNNs
Abstract: Human action understanding and generation is a high-level concept and very hot topic in computer vision. Deep learning methods made significant research progresses in this area. In this task, we aim to understand and generate realistic human actions from a given video input. We will use adversarial domain adaptation algorithm [2] to train Recurrent Neural Networks [1]. More specifically, Generative Adversarial Networks(GANs) will be used to reduce discrepancy between training the network and sampling from the it over multiple time steps. We will compare the accuracy and efficiency of this approach with other recent object detection methods. Literature: [1] Hochreiter, Sepp, and Jürgen Schmidhuber. "Long short-term memory." Neural computation 9.8 (1997): 1735-1780. [2] Goodfellow, Ian, et al. "Generative adversarial nets." Advances in neural information processing systems. 2014. [3] Lamb, Alex M., et al. "Professor forcing: A new algorithm for training recurrent networks."Advances In Neural Information Processing Systems. 2016.
Student:  
Director: Prof. Nassir Navab
Supervisor: Huseyin Coskun, Christian Rupprecht, Federico Tombari
Type: DA/MA/BA
Area: Machine Learning, Computer Vision
Status: finished
Start:  
Finish:  
Thesis (optional):  
Picture:  


Edit | Attach | Refresh | Diffs | More | Revision r1.6 - 24 Mar 2018 - 23:46 - HuseyinCoskun