This work aims to estimate the 6-Degrees of Freedom Pose of an object using simple convolutional neural networks. The problem is that most methods require previous knowledge of the 3D model of the object of interest, which could be unobtainable. We mitigate the problem by simplifying the object’s 3D model to a single and generic primitive solid to create a model that could estimate the pose of unknown objects. Besides that, we study the interpretability of the neural network by using visualization techniques to understand how the network is splitting the high-dimension feature space to reach a Pose estimation.