In order to understand the issue at hand and therefor make it easier to solve we first discus differences and similarities within different computer vision problems so we can make it easier to place the complexity of our semantic segmentation problem that covers classification and localization within this field. We proceed to properly show and visualize our problem so we can understand what exactly are we trying to achieve with given information, how do we get the model to output probabilities for each class for every pixel on our image, and how do we make our data compatible for those nets with categorical shape. By analyzing different convolutional nets, we choose a single architecture that fits our needs and proceed to train it with explained loss function and then evaluate it with different metrics like IoU and categorical accuracy. A simple demonstration of Confusion matrix is also shown so the results are perfectly clear. On every step of the way a Keras and python methods usage is shown so it is easier to follow along. Results are shown with room left to improve and explore. This paper makes a good start for all those interested in the field of computer vision.