In this presentation, we demonstrate the challenges of validating a computer vision (CV) pipeline that runs inside the car and discuss how the task of validating such a pipeline can be solved by using Artificial Intelligence outside the car. It is too expensive to manually label enough reference (ground truth) data for validating the CV pipeline inside the car. One solution is to train a more powerful CV Deep Learning system that runs outside the car on more powerful GPUs and not necessarily in real-time and can provide reference object detections. The latter are compared to the results the algorithm inside the vehicle would produce. The reference deep learning system needs training data to be trained upon. The process of acquiring this initial training data can be optimized by semi-automatic annotation.