EN.601.783

Vision as Bayesian Inference

Tues/Thurs: 9:00-10:15 Spring 2020, Olin 305.


Course Description

This course models vision as Bayesian Inference. It concentrates on visual tasks such as segmenting images, detecting objects in images, and recognizing objects. Its goal is to describe the state of the art techniques. The handouts consist of copies of the lecture notes and related papers.

Reading Material

Homework and Grading Plan

Homework 1
Homework 2
Homework 3
Homework 4

Tentative Schedule

Lecture Topics Handouts Supplements Additional Readings
1 Introduction
Lecture1
2
Image Representation and PCA Sparsity Lecture2
Background Material
Sparse Coding
Eigenfaces
3 Dictionaries, Mixtures of Gaussians and Miniā€Epitomes Lecture3
Math Notes
K-means++
Mini-Epitomes
4 Super Pixels and EM Lecture4: EM and Affinity
Lecture4: Super Pixels
ProtoObjects
SLIC
5 Image Statistics and Weak Membrane Models Lecture5 Part1
Lecture5 Part2

NonLinearTotalVariation
StatisticsImagePatches
6
Edge Detection and Simple Semantic Segmentation Lecture6 Part1
Lecture6 Part2
Math Notes
Linear Filtering
EdgeDetection
SemanticSegmentation
7 Decision Theory Lecture7 Part1
Lecture7 Part2
Lecture7 Part3

8 Deep Networks and Edge Detection Lecture8 Part1
Lecture8 Part2
Lecture8 Part3

Math Details

Holistically-Nested Edge Detection
9 MRF-MFT and Semantic Segmentation Lecture9 Part1
Lecture9 Part2
Lecture9 Part3
DeepLab
Fully Connected VRF
10 Weak Membrane, MRF and Annealing Lecture10
Image Segmentation
Belief Propagation and MFT
11 GrabCut and Belief Propagation Lecture11 Part1
Lecture11 Part2
Lecture11 Part3
GrabCut
CPMC
12 Probabilities on Graphs Lecture12
Bayesian Inference
Neural Implementation
13 Stereo and Boltzmann Machine Lecture13: DP and Stereo
Lecture13: Boltzmann Machine
Bayesian Stereo
Occlusions and Binocular Stereo
Stereo_BP
Stereo_CNN
14 Learning Exponential Models Lecture14: Learning Exponential Models
Lecture14: EM
Inducing Features of Random Fields
FRAME
15 Hidden Markov Models Lecture15
Extract Highlights
16 Motion Lecture16
Math Details
Unsupervised BlackAnandanOpticalFlow
HornShunck80 RobustPointMatching
Motion coherence
17 Geometry and Motion Lecture17 Part1
Lecture17 Part2
Manhattan World Two Algorithms
Factorization methods Symmetry
18 Lighting Lecture 18: LambertianLighting
Lecture 18: Basri1
Lecture 18: Basri2
Lambertian Reflectance and Linear Subspaces
GBR
J53YuilleSnowEpsteinBelhumeur99
19 Adaboost Lecture 19 Part1
Lecture 19 Part2
Notes
ChenYuille
ViolaJonesAdaBoost
20 Deformable Parts Models and SVM Lecture 20: Deformable Part Models
Lecture 20: Support Vector Machines
strang_nonlinear_optimization YuilleHe2013
Latent Support Vector Example
21 Compositional Models Lecture 21 JMIVyuilleB PGMM RCM10cvprLeoZhu
22 Deep Networks Attacks and Understanding Lecture 22: Adversarial Defense
Lecture 22: Understanding Part1 Lecture 22: Understanding Part2
23 Detecting and Parsing Humans; Compositional CNN Lecture23: ParsingHumans
Lecture23: Compositional Convolutional Neural Networks
XianjieChenHumanParsing2014
AdamKortylewskiComPnet2020
24 Computer Graphics and Computer Vision Lecture 24 Part1 Lecture 24 Part2
Lecture 24 Part3
25 Beyond Standard Performance Evaluation Lecture 25 Part1 Lecture 25 Part2
Lecture 25 Part3