Add to ORKGIn a number of fields, neural networks can achieve state-of-the-art performance, but understanding how and why they arrive with a solution is still unclear. Particularly for processing visual data, Convolutional Neural Networks (CNNs) have demonstrated great success. CNNs are structured to function with a two-dimensional image input, and as a result, they maintain the spatial relationships for what the model learns. The main goal of this thesis is to develop a tool for visualization of a CNN's inner activations. We apply different techniques to compare various inputs to provide an educational analysis of CNN's behavior. 1Neuronové siete dosahujú nesmierne výsledky v mnohých oblastiach, avšak stále nie je úplne jasné ako a prečo prídu k dané...