ABOUT THE COURSE: This course explores both classical and deep learning-based approaches to computer vision. Starting from introduction to deep learning, it goes on to discuss traditional approaches as well as deep networks for a variety of vision tasks including low-level vision, 3D geometry, mid-level vision and high-level vision.PREREQUISITES: Familiarity with image processing, linear algebra and probability is desirable but is not a must. INDUSTRY SUPPORT: Google, Amazon, Facebook, Qualcomm, TI, KLA-Tencor, Siemens, GE, Philips etc
Overview
Syllabus
Week 1: Course introduction, Introduction to deep learning, Introduction to neuronWeek 2:Multilayer perceptron (MLP), Gradient descent, Backpropagation in MLPWeek 3:Optimization and regularization, Regularization and preprocessing, Convolutional neural network (CNN)Week 4:CNN properties, CNN architectures, Introduction to recurrent neural network (RNN), Encoder-Decoder models in RNNWeek 5:Low-level vision, Spatial and frequency domain filtering, Edge detectionWeek 6:Line detection, Feature detectors, Harris corner detectorWeek 7:Blob detection, SIFT, Feature descriptors, SURFWeek 8:Single-view geometry, 2D Geometric transformations, Camera intrinsics and extrinsicsWeek 9:Two-view stereo, Algebraic representation of epipolar geometry, Fundamental matrix computationWeek 10:Structure from motion, Batch processing in SFM, Dense 3D reconstructionWeek 11:Deepnets for stereo and SFM, Mid-level vision, Image segmentationWeek 12:Deepnets for segmentation, High-level vision, Deepnets for object detection
Taught by
Prof. A.N. Rajagopalan
