Vehicle orientation detection based on synthetic data with image-to-image translation and domain adaptation
Nguyen Thanh Thien · 2022 IEEE International Conference on Big Data (Big Data) · 2022
In this paper, the author describes the solution for the IEEE BigData Cup 2022 Challenge: Vehicle class and orientation detection in the real-world using synthetic images from driving simulators. This challenge’s task is to enhance the performance of the vehicle detection models, which are trained solely on photo-realistic images, in real-world scenarios. By applying image-to-image translation and domain adaptation technique, the proposed solution can improve the detection results on real-world images using models trained on synthetic data. Furthermore, the proposed method ranked 2nd in the leaderboard with final score of 0.4456.