Tool for automatic labeling of objects in images obtained from Carla autonomous driving simulator

To successfully train modern object detectors in a supervised manner, a large number of labeled images is usually required. Collecting and annotating images can be an expensive and time-consuming job, especially in the field of autonomous driving. A cheaper and faster alternative can be found in computer simulations of real-world traffic scenes, where the object of interest can be automatically labeled. In this spirit, a tool for automatic labeling of the images obtained from CARLA autonomous driving simulator is proposed. The tool runs in parallel with CARLA simulator and creates a synthetic dataset with annotations in the appropriate format for the following objects of interest: traffic lights, traffic signs, vehicles, and pedestrians. The tool enables the end-user to generate a synthetic dataset with a desired number of images and with desired parameters such as weather condition, image resolution, and traffic density. In that way, large synthetic datasets can be generated in a short period of time.