Introduction to Deep Learning and Pattern Recognition for Computer Vision

12 units · Graduate

Fall: 12 units Introduction to Deep Learning and Pattern Recognition for Computer Vision will focus on Deep Learning algorithms used in Computer Vision applications while explaining the pattern recognition aspect of these algorithms. The first half of this course is also available as a mini ( 18-790 ) which introduces students to the basic Deep Learning ML techniques in the course. The course will first introduce Neural networks and how they perform recognition and their evolution to Deep Neural Networks, as well as different DNN backbone architectures (e.g. VGG, ResNet + variations, MobileNets, etc.) used for classification. We will overview DL architectures for object detection (to include a large range of algorithms such as anchor-based and anchor free, single stage, two-stage as well as well-known Yolo, SSD, FCOS, CornerNet, Mask-RCNN, DETR and others). We cover object recognition, semantic segmentation (with applications in robot vision, autonomous driving, general scene understanding, medical analysis), and other topics including instance segmentation, loss functions, feature extraction, Transformers, Generative Models, Neural Architecture Search (NAS), low form factor Deep Learning architectures for embedded platforms (e.g., Jetson Nano, AGX), TensorRT for model optimization on Nvidia embedded platforms, and ONNX model conversions. Prerequisite: 36-219 Min. grade C