2021/01/13 by Anton Konushin, Konushin, Anton, Boris Faizov +3
Computer Science · #Advanced Image Processing Techniques #Computer Vision and Pattern Recognition (cs.CV) #Digital Media Forensic Detection #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Image Processing and 3D Reconstruction #Image and Object Detection Techniques
paper · pdf · doi:10.48550/arxiv.2101.04927
openalex publication_date 2021/01/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Traffic sign recognition is a well-researched problem in computer vision.\nHowever, the state of the art methods works only for frequent sign classes,\nwhich are well represented in training datasets. We consider the task of rare\ntraffic sign detection and classification. We aim to solve that problem by\nusing synthetic training data. Such training data is obtained by embedding\nsynthetic images of signs in the real photos. We propose three methods for\nmaking synthetic signs consistent with a scene in appearance. These methods are\nbased on modern generative adversarial network (GAN) architectures. Our\nproposed methods allow realistic embedding of rare traffic sign classes that\nare absent in the training set. We adapt a variational autoencoder for sampling\nplausible locations of new traffic signs in images. We demonstrate that using a\nmixture of our synthetic data with real data improves the accuracy of both\nclassifier and detector.\n