Object detection is a critical task in today's world, with applications spanning many diverse fields, including, but not limited to autonomous vehicles, biometric and facial recognition, and industrial automation. Traditional object detection methods heavily rely on ample amounts of labeled training data and require extensive training time. This makes them resource-intensive and less adaptable to rapidly changing scenarios. In this thesis, we define a one-shot object detection framework that overcomes these limitations. Using state-of-the-art pretrained transformer networks, our approach enables real-time detection of novel objects from a single reference image, eliminating the need for a training phase. This results in a scalable and efficient solution for dynamic, data-scarce environments.