Master Thesis: Optimization of transfer learning methods to simplify the change of image sources for neural networks

Intern / Student, Part-time ยท Munich

Your tasks
Deep neural networks (CNNs) are trained on images with specific attributes like resolution (GSD), color distribution, channels (RGB, RGBI, spectral, etc.). They deliver optimum results on the imagery they have been trained on. If images are changed for the detection process, the quality of detection usually decreases.

Objective is to reduce the required efforts for transfer learning when changing the image source for a trained neural network.
Requirements
Skillsets:
  • Machine learning (CNN, deep learning) training and inference
  • Programming with Python
  • Use of image libraries like GDAL etc.
  • Visualization with GIS tools
Areas of expertise:
  • Data augmentation
  • Modification of neural network architecture used
  • Methods to harmonize imagery with regards to their machine learning relevant attributes (up-/downsampling, morphing, GANs)
Life as TerraLouper is
  • exciting to be part of a highly motivated, international top team of engineers
  • interesting to work with cutting-edge technologies, agile development practices and product-focused environment
  • office location in the heart of Munich
  • flexible working hours
  • multi-cultural team
About us
TerraLoupe is a German startup to empower autonomous mobility. We use machine learning technology to automaticcally detect objects from aerial imagery to create a 3D digital twin of the world. Located in the heart of Munich the next level of geo platform and HD mapping is getting to reality.
Your application
Thank you for considering a career at TerraLoupe GmbH. Please fill out the following form. In case you are experiencing problems with the document upload, mail your documents to careers@terraloupe.com.

Please upload any documents that you want to include with your application. CV and cover letter are required, and you should also attach copies of your references and certificates.

Click to select multiple files or use drag-and-drop
Uploading document. Please wait.