Multilingual scene text detection and recognition are active research problems. Distinguishing text from non-text objects and, thereafter, identifying the script of the text are important tasks in this notion. The advent of deep learning and its unprecedented performance have led to the development of many architectures and variants of its baseline network called the convolutional neural network. In this chapter, seven relevant architectures namely, LeNet, GoogLeNet, Visual Geometry Group (VGG) 16, Xception, MobileNet, InceptionV2, residual attention network, and ResNet-152 have been applied for the above tasks on manifold sets of images. The obtained performances revealed that a few of these architectures are really high performing, for example, ResNet-152 is found to be the most suitable for text non-text image classification, while VGG16 is its nearest contestant. On the other hand, GoogLeNet, ResAttNet, and ResNet-152 appear competitive for script identification. Furthermore, their performances beat the state of the art by a considerable margin.