/ComputerVisionCourse

OtherNOASSERTION

Computer Vision Course

Module 1. Image processing basics. Local image processing and feature descriptors

Basic image processing:

  • Scale pyramid, bluring, Image gradients, Contour detection.

Correspondences and wide baseline stereo:

  • Interest point and distinguished regions detection:
  • Harris operator (corner detection)
  • Hessian detector, affine covariant version, Maximally Stable Extrema Regions (MSER).
  • Descriptors of measurement regions
  • SIFT (scale invariant feature transform), RootSIFT
  • Shape context.
  • LBP (local binary patterns)
  • Deep learned features (HardNet).
  • Deep learned features (R2D2, SuperPoint)

Graph neural networks for matching (LoFTR, SuperGlue)

Robust model fitting:

  • RANSAC.
  • Hough transform

Other:

Module 2. Image Segmentation and Correspondence

Intro into segmentation, matting and correspondence problems

Image segmentation:

  • Level-set methods
  • Otsu thresholding
  • Postprosessing (erosion, dilation, NL-means)
  • Edge-based methods
  • Connected components labeling
  • Active contour model
  • Super-pixel segmentation
  • Watershed segmentation
  • Expectation-maximization in computer vision
  • Gaussian mixture model and K-means
  • Segmentation based on graph cuts
  • Markov models (chain, trees, P2D, CRF)
  • YOLACT

Image correspondence:

  • Sparse and dense optical flow
  • Affine and projective transformations
  • Birchfield–Tomasi dissimilarity
  • Dynamic programming for stereo correspondence
  • Polar coordinates and 360 photo
  • Panorama stitching
  • Lukas-Kanade Optical Flow
  • Horn-Schunck optical flow
  • Gunnar-Farneback algorithm
  • PatchMatch algorithm
  • PWC-Net, MaskFlownet

Module 3. Object Tracking and Search

  • Histograms and statistical models
  • Hidden Markov Models
  • Integral images
  • HOG detector
  • KLT tracker
  • Mean-Shift, CamShift tracker
  • Kalman Filter
  • Binary features
  • Bag of visual words
  • minHash
  • Image Retrieval for large image collections: image description, indexing, geometric consistency
  • Neural nets for object tracking and search

Module 4. Geometry and Augmented Reality (AR)

  • Homogeneous coordinates
  • Translation, rotation, scale, and projection matrices
  • Camera models
  • Rendering 3D scenes
  • Camera calibration and removal of lens distortion
  • Search for lines, circles and ellipses in photo
  • Stereo and epipolar geometry
  • Match moving and 3D reconstruction
  • Light and materials
  • AR examples

Module 5. Deep Learning for CV

  • ML basics: data retrieval, synthesis, augmentation, train and test sets
  • Non-deep ML models (PCA, SVM)
  • Deep learning
  • Different layers. Math and implementation.
  • Optimization for size and speed (MobileNet, ... )
  • Running in real-time (Core ML 2, etc)
  • Study of different architectures: Image classification (AlexNet), Object detection (R-CNN), Object Tracking (SAE and CNN), Object segmentation (SegNet), Instance segmentation (Mask R-CNN), Optical flow (PWC Net), Human pose estimation (VNect), 3D reconstruction (LayoutNet), Transfer-based AR (LSTM Neural nets, GANs). Metrics learning (face recognition)