|
Computer Vision (MA-INF 2201) Summer Semester 2010
Times:
Monday & Wednesday: 9 - 11am (lectures); Thursday: 8:15 - 10 am (exercises) Place:
rooms A207 (lectures), and A7a (exercises), both in Romerstr. 164 Syllabus
(please check regularly for course
materials and updates!) No classes during
the week of May 24th-28th (Pentecost Break) To access
course materials, get password in class or send email to:
tutors.inf2201.10 at gmail.com
(provide your student ID) Assignments or
exams can be turned-in/taken in English or in German Subject Readings Assignments
1 Apr.12th: Introduction Course logistics: pre-requisites,
resources, programming and exercises. Apr.14th: Overview of computer
vision, problems and challenges Introduction.
An overview of what can we do today with computer vision. Typical vision
sub-problems. Why is vision hard? Challenges and open issues.
Slides:
2 per page.
4 per page. 2 Apr.19th: Image representation
and filtering Image representation. Perception and
spatial scale. Linear filtering, correlation and convolution
operators. Noise and the calculation of image gradients. Edge preserving
(bilateral) filtering.
Slides:
2 per page.
4 per page.
Forsyth and Ponce, Ch. 7.1,7.2, 8.1, 8.2 Apr.21st: Edge detection Edge detection.
Laplacian operators. Sampling and aliasing. Fourier analysis. Image
pyramids.
Slides:
2 per page.
4 per page.
Forsyth and Ponce, Ch. 7, 8 3 Apr. 26th: Interest point
detectors Pyramid blending. Image feature
detectors. Auto-correlation functions and mathematical analysis for the
Förstner-Harris
operator. Scale-invariant extensions.
Slides:
2 per page.
4 per page. Apr.28th: Descriptors and matching
Affine-invariant interest point
operators. SIFT and Shape Context feature descriptors. Distinctiveness and
repeatability. Constrained global image correspondence problems.
Linear assignment and the Hungarian algorithm.
Slides:
2 per page.
4 per page. 4 May 3rd: Projective geometry for
computer vision Projective plane, projective
transformations and projective invariants. Line and point duality. Relation
between projective, affine and euclidean transformations. Applications of
cross-ratio.
Slides:
2 per page.
4 per page. Hartley and
Zisserman, Ch. 2.1 - 2.4, 3.1-3.2.2 May 5rth: Geometric Constraints
and the Fundamental Matrix Camera models. Computation of
homographies and camera matrices from point correspondeces. Epipolar
geometry and the fundamental matrix. Algebraic and statistical methods.
Estimation and normalization.
Slides:
2 per page.
4 per page. 5 May 10th: Stereo The essential matrix. Orthographic
factorization methods. Rectification. Disparity-based representations, the
cyclopean eye and coherency-based stereo. Stereo constraints. Plane sweeping
methods.
Slides:
2 per page.
4 per page. May 12th: Optical Flow Image warping. The motion field and
the optical flow. Brightness constancy equations and first order
approximations. Translational (Lucas-Kanade) optical flow. Affine flow.
Hierarchical estimation methods. Layered models.
Slides:
2 per page.
4 per page. 6 May 17th: Lighting and color Color constancy and the physics of
light. Radiometry, BRDF models. Low-dimensional models for color spectra.
Color matching and Grassman's laws. Measuring color by color matching.
Slides:
2 per page.
4 per page. May 19th: No lecture, "Dies Academicus". 7 May 31st: Image segmentation I.
Modeling and clustering Visual grouping and Gestalt
principles. Image segmentation as a clustering problem. Agglomerative and
divisive methods, K-means and mean shift. Spectral graph partitioning
methods.
Slides:
2 per page.
4 per page. Jun 2nd: Image segmentation II.
Contours and snakes Hough transform. Contour
segmentation methods. Snakes and live-wire. Discrete and continuous
formulations. Elasticity and stiffness priors. Gradient flow and dynamic
programming solvers.
Slides:
2 per page.
4 per page. 8 Jun 7th: Machine learning for
computer vision The role of machine learning in
computer vision. Hypothesis spaces, generalization and overfitting.
Generative and discriminative models. Dependencies on graphs,
conditional and marginal parameterziations. d-separation and the Bayes ball
algorithm.
Slides:
2 per page.
4 per page. Jun 9th: Support Vector Machines The bias-variance decomposition.
Linear maximum margin classifiers. Primal and dual formulations. Support
Vector Machines. Feature spaces and kernels. Mercer's condition and kernel
closure rules. Kernelizing algorithms.
Slides:
2 per page.
4 per page. 9 Jun 14th: Learning for object
recognition I Introduction to object recognition.
Eigenfaces: face recognition using Principal Component Analysis (PCA).
K-means and bag-of-feature models.
Slides:
2 per page.
4 per page. Jun 16th: Learning for object
recognition II Processing pipeling in the visual cortex. Hierarchical feature extraction architectures,
based on template matching and max-pooling operations: HMAX by Riesenhuber and Poggio '99, extensions by Mutch and Lowe '08.
Slides:
2 per page.
4 per page.
Assignment 5: helper function to draw a descriptor.
Receive Assignment 6 10 Jun 21st: Boosting and sliding
window methods for object detection Ensemble classifiers and
meta-learning frameworks. Bagging, stacking and boosting. Functional
gradient descent. Adaboost and
sliding-window methods for object detection.
Slides:
2 per page.
4 per page.
Jun 23rd: Branch and bound optimization for
efficient image search.
Slides:
2 per page. 4 per page.
11 Jun 28th: Learning in fully
observed statistical models Regression. Feature selection
methods, shrinkage and the lasso. Bayesian classifiers and linear
discriminant analysis. Naive Bayes.
Slides:
2 per page.
4 per page.
Jun 30th: Latent variable models
and the EM algorithm Learning in partially observed models. Mixture of
experts. Complete and incomplete log-likelihoods. Free energy and view of EM
as bound optimization. Illustration for Gaussian mixture fitting.
Slides:
2 per page.
4 per page.
12 Jul 5th: Dimensionality reduction
and continuous latent variable models Linear (PCA) and non-linear
dimensionality reduction methods (ISOMAP, Laplacian Eigenmaps). Factor
Analysis and Probabilistic PCA. Spectral latent variable models.
Applications to visual inference (object tracking, 3d reconstruction).
Slides:
2 per page.
4 per page.
Jul 7th: Belief Propagation Elimination methods. Moralization
and triangulation. Belief propagation in tree-structured models. Hidden
Markov Models.
Slides:
2 per page.
4 per page. 13 Jul 12th: Learning random fields. Log linear models and maximum entropy.
Slides:
2 per page.
4 per page.
Jul 14th: Reconstruction of 3D articulated objects. Levels of 3D modeling. State space priors and observation models. Generative vs. discriminative models. Particle filters.
Slides:
2 per page.
4 per page.
14 July 19th: Optimization and sampling methods. High-dimensional visual inference. Local maxima and transition states. High-dimensional sampling strategies (hyperdynamics, hypersurface sweeping).
Slides:
2 per page.
4 per page.
July 21st: Exam |