This course aims at providing a basic understanding of the various kinds of vision models in deep learning for computer vision. It initiates by illustrating the methods used for focusing on a specific feature of an image in a large volume of datasets. You will discover the role of computer vision techniques in generating textual content for an image. Next, you will study the relevance of attention models in performing tasks such as visual question answering and dialogues. You will explore the procedure for localising the vital parts of an image using spatial transformer networks. The course explains the various forms of attention mechanisms that emphasise hidden states and parallelisation. In addition to this, we will describe the overall perspectives of deep generative models in computer vision. You will explore the significance of generative adversarial networks in translating information from visual content. Explore the methods of ascertaining the probability of patterns in the perpetual space.

Next, the course illustrates the various methods of combining generative adversarial networks (GANs) and variational autoencoders (VAEs) in a single framework. You will discover the process of modelling and transforming complex densities using non-linear independent components estimation and real-valued non-volume preserving methods. Following this, we investigate the variations of generative adversarial networks. You will discover how to perform image translations and embed latent spaces to obtain diverse images. Subsequently, you will study the methods of encoding closely distinct variables as discrete dimensions using disentangled representation. Next, we outline the applications of adversarial networks and generative models concerning images and videos. You will study the process of estimating accurate classes with and without exposure to any occurrences of the training datasets. Then, the course explores the notion of self-supervised learning in computer vision. You will discover the process of predicting a particular part of the input based on other parts of the data.

Finally, the course illustrates the various methods of estimating adversarial robustness. This method will include the procedures for tackling the malfunctions of learning models using different adversarial defence mechanisms. In addition to this, you will notice how pruning and quantisation minimise irrelevant parameters that do not influence performance. Lastly, you will study the concept of neural architecture search in deep learning. This notion encompasses the methods of searching for the right neural network architecture for a given problem. Study the recent advances in deep learning for computer vision, emphasising topics such as semantic image segmentation, multi-model learning for organised label spaces, adversarial robustness, deep model compression and design of artificial neural networks (ANN). The ‘Diploma in Models and Trends in Computer Vision’ is an informative course that illustrates the recent success of deep learning methods in revolutionising computer vision making new developments increasingly closer to deployment that benefits the end-users.