Road images augmentation with synthetic traffic signs using neural networks

Traffic sign recognition is a well-researched problem in computer vision. However, the state of art methods works only for frequent classes, which are well represented training datasets. We consider task rare traffic detection and classification. aim to solve that by using synthetic data. Such data obtained embedding images signs real photos. propose three making consistent with scene appearance. These based on modern generative adversarial network (GAN) architectures. Our proposed allow realistic classes absent set. adapt variational autoencoder sampling plausible locations new images. demonstrate mixture our improves accuracy both classifier detector.