One of the problems in training neural networks models is the
availability of properly labeled datasets. In particular in the areas of 6d
pose recognition and in general for 3d object recognition there are not
many datasets available.  

In this master thesis you will use a photogrammetry algorithm to
recreate 3D models of real objects and then through simulation
environment created in Unity generate depth images and the
corresponding labels in an automated way.  

Optionally (for a Master thesis) a set of benchmarks could be created
on different type of objects (e.g. featureless, transparent, reflective)
and then applied to existing neural networks for 6d pose recognition.
During this thesis you will have the opportunity to work with machine
learning experts and learn machine learning computer vision and
robotics.

Mandatory:
- Some programming experience
Optional:
- Familiarity with Unity tools and scripting
- Python programming
- Experience with Tensorflow