Transparent and metallic objects are common in our modern world. However, due to their inherent properties, they pose significant challenges for computer vision algorithms. In this project we aim to improve the most promising existing state-of-the-art pose estimation models for transparent and metallic objects regarding their execution time. The goal is to conceptualize, implement and evaluate a network, which is capable of estimating poses of those challenging objects on a notebook (or smartphone) without diminished capabilities.