A seminar report submitted by nidish kumar r V, ra1911003020205

We can also analyze the characteristics of the dataset firstly, to conduct data augmentation more effective. The distribution of the training set is not uniform and traffic signs of some classes are too few, which will influence the recognition ability of the network. In this paper, we will generate two datasets called the Equaled Set and the Enlarged Set through data augmentation. Equaled Set: This set is balanced across classes using data augmentation. Each class has 30,000 samples, which contain the original training set. Enlarged Set: This set which is expanded through data augmentation contains 30× more data than the original one. The data augmentation technologies we used included the rotation, reflection, and horizontal/vertical flip.