Explore fundamental computer vision tasks and neural network architectures in this comprehensive lecture from JSALT 2025. Learn about key computer vision challenges and discover the neural network models commonly used to solve them, with detailed coverage of model architectures and training methodologies. Dive deep into large Transformer-based models and examine contemporary pre-training approaches that have revolutionized the field. Investigate handwritten text recognition systems and understand their connections to speech recognition techniques, drawing parallels between these related domains. Gain insights from Michal Hradiš of Brno University of Technology, who brings extensive experience as the principal developer of PERO OCR and expertise in semantic document understanding, information retrieval, and medical computer vision applications.