Take a look at the advertised positions below if you would like to get involved in data science projects or join the Data Science Unit team.
Masterthesis: Rotation invariance in deep learning for underwater visual recognition
When photographing the seafloor from an underwater robot, objects can appear variously rotated in the images. Finding back the same spot on the seafloor requires more careful handling of camera geometry than in air, where most images are aligned to gravity direction. Therefore, neural networks that are pretrained on gravity-aligned datasets struggle in seafloor mapping applications. A potential solution could be to equip convolutional neural networks with rotation invariance properties.
To adress this challenge, we offer a master thesis topic about different variants to include rotation and scale invariance into common architectures of convolutional neural networks in order to increase robustness of learned feature matching methods. (Flyer download)