Description: [SCS dragon logo]

 


CS280: Computer Vision
 
Computer Science Division
University of California Berkeley

INSTRUCTOR: Jitendra Malik
GSI: Pulkit Agrawal
GSI: Yuansi Chen 
UNITS: 3
SEMESTER: Spring 2015

COURSE OVERVIEW

Computer vision seeks to develop algorithms that replicate one of the most amazing capabilities ofthe human brain – inferring properties of the external world purely by means of the light reflectedfrom various objects to the eyes. We can determine how far away these objects are, how they areoriented with respect to us, and in relationship to various other objects. We reliably guess theircolors and textures, and we can recognize them - this is a chair, this is my dog Fido, this is a pictureof Bill Clinton smiling. We can segment out regions of space corresponding to particular objectsand track them over time, such as a basketball player weaving through the court.

In this course, we will study the concepts and algorithms behind some of the remarkable suc-cesses of computer vision – capabilities such as face detection, handwritten digit recognition, re-constructing three-dimensional models of cities, automated monitoring of activities, segmentingout organs or tissues in biological images, and sensing for control of robots. We will build thisup from fundamentals – an understanding of the geometry and radiometry of image formation,core image processing operations, as well as tools from statistical machine learning. On completingthis course a student would understand the key ideas behind the leading techniques for the mainproblems of computer vision - reconstruction, recognition and segmentation – and have a sense ofwhat computers today can or can not do.
 

TOPICS TO BE COVERED

  • Introduction - The Three R's - Recognition, Reconstruction, Reorganization
  • Static Perspective - the pinhole camera model
  • Transformations - rotation, translation, affine and projective
  • Dynamic perspective and optical flow
  • Radiometry of image formation
  • Basic image processing operations - filters, features and flow
  • Biological visual processing - retina, V1 and beyond
  • The feedforward model of visual processing - convolutional networks
  • Object recognition case study - Identifying digits with multiple approaches
  • Recognizing objects in scene - sliding windows and object proposals.
  • Feature Histograms
  • Convolutional Neural Network (ConvNet) based approaches to visual recognition of objects and scenes
  • Attributes, pose and actions
  • Controur detection and bottom-up segmentation, Gestalt grouping heuristics
  • Semantic Segmentations - instance segmentation and pixel classification
  • 3D reconstruction from multiple views
  • 3D reconstruction from pictorial cues
  • Scene understanding from RGBD images
  • Face Recognition
  • Video Analysis

  

COURSE MATERIAL

Lectures

Material

Lecture 1: Introduction


 

Lecture 2: Fundamentals of Image Formation (Static Perspective)


 

Lecture 3: Transformations


Pinhole Camera

 

 

 

Lecture 4: Dynamic Perspective


 

 

Lecture 5: Radiometry of Image Formation


 

 

Lecture 6: Basic Image Processing


 

Lecture 7: Biological Visual Processing


 

 

Lecture 8: Handwritten Digit Recognition


 

Lecture 9: VIsual Grouping


 

 

Lecture 10: Object Detection Using ConvNets


 

 

Lecture 11: Deformable Parts Model (DPM)


Lecture 12: Binocular Stereopsis


 

Lecture 13: Binocular Stereopsis II


 

 

Lecture 14: Markov Random Fields in Computer Vision


 

 

Lecture 15: Solving for Stereo Correspondence


 

 

Lecture 16: Optical Flow


 

Lecture 17: Action Recognition


 

 

Lecture 18: Simultaneous Detection and Segmentation


auto

 

 

Lecture 19: Pose and Keypoint Estimation


1

 

 

Lecture 20: Review of Differential Geometry


multi-perspective

 

 

Lecture 21: Scene Understanding from RGBD Images


HDR

 

 

Lecture 22: 3D Perception from a Single image


1

 

 

 

Lecture 23: Face Recognition


 

 

Page design courtesy of Doug James