The growing importance of deep learning in the field of computer vision is undeniable and has the potential to change the way we approach image and video analysis. A project focusing on efficient and interactive training of deep learning systems is an important step in this direction. The development of an architecture and training process that utilizes the principles of active and semi-supervised learning allows for purposeful selection of training data and early exploitation of knowledge gains in the training process. With the long-term goal of estimating whether a model will be able to fulfill requirements defined by the use case early in the training process, this method could be a significant breakthrough in computer vision research. The application of active and semi-supervised learning is a promising approach to reduce the required amount of training data and improve the learning speed of deep learning systems. Active learning selectively chooses training data by asking the model which data is most informative to achieve the greatest learning gain. Semi-supervised learning utilizes labeled and unlabeled data to improve the model. The model is encouraged to recognize and generalize patterns in the unlabeled data to make better predictions. Overall, this project is a promising step towards more efficient and interactive training of deep learning systems in computer vision research.