The main goal of this thesis will be to design and train a deep neural network to estimate surrounding illumination from a single RGB image of a human face. The resulting illumination will be estimated in form of an HDR environment map. A large dataset of human face renderings will be created by the Unreal Metahuman Tool using many different environment maps as light sources. This dataset will be then used together with existing image datasets to train the designed model.


  • Study related work about light source estimation
  • Prepare a large dataset of image pairs (source image of a face, resulting HDR environment map)
  • Design and train a deep neural network for light source estimation


  • Knowledge of python
  • Experience with deep learning
  • Experience with game engines is advantageous


  • Tensorflow or PyTorch
  • Unreal Engine
  • Maybe Maya


For more information please contact Peter Kán.



Student Project
Master Thesis