CS484/684
list of topics
This course introduces a number of standard computer vision problems and computational methods for solving them. A tentative list of topics is given below (not necessarily in order they will be taught):
- Common Image Modalities: pin hole camera, photo, video, medical, etc.
- Point Processing and Local Processing (Filtering):
- gamma correction, window/center adjustment, histogram equalization
- linear transforms, gradients, convolution, cross-correlation, non-linear filtering.
- Features and Descriptors: colors, edges, corners, SIFT, MOPS, etc.
- Image warping: domain tranforms, linear/affine, projective, homographies.
- Model fitting/estimation
- color models: non-parametric (histograms, kernel densities), parametric (Gaussian, GMM), ML estimation, segmentation with color models
- geometric models (lines, planes, homographies): panoramas, 3D reconstruction, robust estimation (RANSAC), multi-model fitting.
- Stereo and multi-view reconstruction, Structure-from-Motion
- projection matrices, camera calibration, epipolar geometry, fundamental and essential matrices, disparity maps, optical flows, volumetric shape reconstruction
- from window-based towards regularization-based stereo, loss functions
- Unsupervised and weakly-supervised Segmentation
- basics: thresholding, region growing
- quantization, clustering (K-means, kernel/spectral methods, mean-shift)
- boundary regularization, edge alighnment
- Machine Learning Basics
- linear classifiers, neurons/perceptrons, decision functions (step-function, sigmoid, soft-max)
- training and testing, loss functions, accuracy measures, overfitting, generalization
- non-linear classification, multi-layered neural networks
- Introduction to Convolutional Neural Networks
- architectures for image classification, convolutional layers, activation functions, attention (CAM)
- encoder-decoder architectures for semantic segmentation, depth reconstruction, and other pixel labeling problems
- convolution and deconvolution, pooling, stride, upsampling, skip-connections, field-of-view etc.
- loss functions and optimization (gradient descent, backpropagation)
- network regularization, drop-outs, batch-normalization
- applications: classification, segmentation, monocular depth, NeRF, inpaining, etc.
- full and weak supervision, self-supervision, unsupervised discriminative and generative models, auto-encoders
- deep clustering