I am currently writing my masters thesis about using synthetic videos (generated using game engines) to improve the accuracy of action recognition models.
Stay tuned for more, i will publish first results soon!
Human action recognition in public places is a challenging task. In this thesis a new dataset of actions in public places is presented, which was generated by a 3D simulation. It is then shown that the dataset can be used to improve existing action recognition models.
will be published on https://github.com/AlexanderMelde after the official examination procedure.
what's the stack? did I run into issues or discover something new along the way?
Currently using Unity Game Engine and Keras, more information will be published here after the official examination procedure.
I have the feeling this will lead to a success, stay tuned and follow me here or on GitHub for more information!