Visual Recognition and Search
Rogerio Feris, Jan 23, 2014
EECS 6890 – Topics in Information Processing
Spring 2014, Columbia University http://rogerioferis.com/VisualRecognitionAndSearch2014
Introduction
Instructors:
Rogerio Feris http://rogerioferis.com
Liangliang Cao http://researcher.watson.ibm.co
m/researcher/view.php?person= us-liangliang.cao
Visual Recognition And Search
Jun Wang http://researcher.watson.ibm.co
m/researcher/view.php?person= us-wangjun
Columbia University, Spring 2014
Introduction
Visual Recognition And Search Columbia University, Spring 2014
Introduction
First Digital Camera (1975)
0.01 Megapixels
23 seconds to record a photo to cassette
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Datasets with 5 or 10 images
Large-Scale Experiment: 800 photos (Takeo Kanade Thesis,
1973)
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Announcement of Pope Benedict in 2005
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Announcement of Pope Francis in 2013
Rapid proliferation of mobile devices equipped with cameras
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Era of Big Visual Data
Billions of cell phones equipped with cameras
~500 billion consumer photos are taken each year world-wide
~500 million photos taken per year in NYC alone
Hundreds of millions of Facebook photo uploads per day
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Visual Recognition And Search Columbia University, Spring 2014
Introduction
How to annotate large amounts of visual data?
Crowdsourcing: Amazon Mechanical Turk
ImageNet: Millions of Annotated Images , Thousands of Categories http://www.image-net.org/
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Impressive results obtained by deep convolutional networks
Visual Recognition And Search Columbia University, Spring 2014
Introduction http://horatio.cs.nyu.edu/
Visual Recognition And Search Columbia University, Spring 2014
Introduction
+ DATA
+ Computational Processing
+ Advances in Computer Vision and Machine Learning
Visual Recognition And Search Columbia University, Spring 2014
Introduction
Visual Recognition and Search: Applications & Challenges
Student Introductions
Course Overview and Requirements
Syllabus Tour
Cool Data for Projects
Visual Recognition And Search Columbia University, Spring 2014
Applications
Visual Recognition And Search Columbia University, Spring 2014
Applications
Self-Driving Cars
Object Recognition and
Tracking – mostly 3D point clouds (laser scanner)
Passed Driver’s license test in
Nevada!
Check Sebastian Thrun short TED talk: http://www.ted.com/talks/sebastian_thrun_google_s_driverless_car.html
CVPR 2012 talk with more details: http://techtalks.tv/talks/self-driving-cars/56391/
Visual Recognition And Search Columbia University, Spring 2014
Applications
Driver Assistance http://www.mobileye.com/
Collision Warning
Automatic Head Light Control
Lane Departure Control etc.
Visual Recognition And Search
Check interesting talk by Amnon Shashua: http://www.youtube.com/watch?v=cg5SNoTF3Ms
Columbia University, Spring 2014
Applications
Seeking the strongest rigid detector (CVPR 2013)
Rodrigo Benenson, Markus Mathias, Tinne Tuytelaars, Luc Van Gool
Visual Recognition And Search Columbia University, Spring 2014
Applications
Mars Exploration Rover
Visual Recognition And Search
Autonomous Navigation
Highly textured environment – good for stereo!
Terrain Classification
Automatic Photo capture of interesting scenes
Horizontal velocity estimation in landing – feature tracking
Columbia University, Spring 2014
Applications
Gaming
Kinect Camera: RGB + Depth sensor
Human pose Detection and Gesture Recognition
See J. Shotton et al, Real-Time Human Pose Recognition in Parts from a Single
Depth Image, CVPR 2011 (Best paper award)
Visual Recognition And Search Columbia University, Spring 2014
Watch for Abnormal Events!
Applications
What did you notice in the video?
People walking?
Vehicles driving?
Abandoned bag???
Visual Recognition And Search Columbia University, Spring 2014
Applications
Visual Recognition And Search Columbia University, Spring 2014
Applications
Surveillance http://www.today.com/video/today/51630165/#51630165
Visual Recognition And Search Columbia University, Spring 2014
Applications
People Search in Surveillance Videos
Visual Recognition And Search Columbia University, Spring 2014
Applications
Video Demo: http://rogerioferis.com/demos/PeopleSearch_BlurredFaces.wmv
Visual Recognition And Search Columbia University, Spring 2014
Applications
Boston Bombing Event
1071 detected faces from 50 high-res Boston images (all from Flickr)
Ability to spot a person with e.g., a white hat in a crowded scene
Suspect #1 found in 4 images in top 8 results Suspect #2 found in 3 images in top page
Visual Recognition And Search Columbia University, Spring 2014
Applications
Visual Recognition And Search
Cashier Fraud Detection:
Faked Scans (aka
Sweethearting fraud)
Giving-away of merchandise without charge to a "sweetheart" customer (e.g., friend, family, fellow employee) by faked product scan
Q. Fan et al, Recognition of
Repetitive Sequential Human
Activity, CVPR 2009
Columbia University, Spring 2014
Like a normal field guide…
that you can search and sort
and with visual recognition
See N. Kumar et al,
"Leafsnap: A Computer
Vision System for
Automatic Plant Species
Identification, ECCV 2012
Nearly 1 million downloads
40k new users per month
100k active users
1.7 million images taken
100k new images/month
100k users with > 5 images
Users from all over the world
Botanists, educators, kids, hobbyists, photographers, …
Slide Credit: Neeraj Kumar
Applications
Google Goggles
Amazon
Visual Recognition And Search Columbia University, Spring 2014
Applications
Auto-focus by Face Detection http://www.rogerioferis.com/demos/Face
TrackerRealTime.avi
TAAZ Virtual Makeover http://www.taaz.com/
Visual Recognition And Search Columbia University, Spring 2014
Applications
IntraFace Demo
(http://www.humansensing.cs.cmu.edu/intraface/)
Visual Recognition And Search Columbia University, Spring 2014
Applications
Finding Visually Similar Objects
Visual Recognition And Search Columbia University, Spring 2014
Applications
Exploring Community Photo Collections http://phototour.cs.washington.edu/
Visual Recognition And Search Columbia University, Spring 2014
Applications
Organizing Photo Collections – IBM IMARS http://researcher.watson.ibm.com/researcher/view_project.php?id=877
Visual Recognition And Search Columbia University, Spring 2014
Applications
Many More Applications…
Medical Imaging
Biometrics
Handwritten Digit Recognition
And so on ….
Visual Recognition And Search Columbia University, Spring 2014
Challenges
Visual Recognition And Search Columbia University, Spring 2014
Challenges
Visual Recognition And Search
Slide Credit: Kristen Grauman
Columbia University, Spring 2014
Break
Visual Recognition And Search Columbia University, Spring 2014
Student Introductions
Student Info Form (see website: Homework #0)
Send by Monday, Jan 27
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Understand and analyze the state-of-the-art in visual recognition and search.
Hands-on experience with cutting-edge methods for visual recognition and search.
Note: 7:00pm – 8:50pm (3 credits)
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
1) Courses/Background in Computer Vision or
Image Processing is mandatory.
2) Programming skills in C/C++ or Matlab are also required.
Please drop the class if you don’t satisfy these requirements
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Each class will consist of:
Lecture by instructor
Student presentations and discussion
Paper Reviews
Solo (no groups)
Project
Final project report: Paper (4-8 pages)
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Class Participation (10%)
Paper Presentations (20%)
Paper Reviews (20%)
Term Project (50%)
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Check course webpage for:
Publicly Available Source Code
Datasets
Related Courses
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Students will be divided in groups and will have to prepare one or two presentations in class (list of presentation topics will be provided soon).
More information about the format and length/time of the presentations will be available in the course webpage.
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Presentation slides will have to be sent to instructors at least one day before the presentation.
Grading based on clarity, organization, knowledge, and whether the presentation meets the time constraints.
Check http://www.slideshare.net/cameraculture/how-to-givea-good-talk
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Individual evaluation (no groups)
Each review (1-2 pages) should address paper strengths, weaknesses, quality of experiments, and describe ways in which the paper could be extended.
Check the course webpage for more details
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Groups of two or three depending on the total number of students attending the class.
Students are expected to spend significant time on their projects
Project Proposal, Project Update I, Project Update II, Final
Presentation, and Final Project Report (4-8 pages)
Visual Recognition And Search Columbia University, Spring 2014
Course Overview
Projects will be graded according to the following criteria:
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Broad set of topics; focus on the state-of-the-art
Check the schedule and required reading for each class in the course webpage
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Low-Level Representation: Feature
Detection and Description
Classical Descriptors: SIFT and SURF
Modern Descriptors for Real-Time Applications: FAST, BRISK, …
Check ECCV 2012 Tutorial on Modern Features: https://sites.google.com/site/eccv12features/slides
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Mid-level feature coding and pooling
Classical Bag-of-Word Models
Spatial Pyramid Matching
Visual Recognition And Search
Modern higher-level representations:
Fisher vector and super-vector encoding, LLC, …
[K. Chatfield et al, The devil is in the details: an evaluation of recent feature encoding methods, BMVC 2011]
Columbia University, Spring 2014
Syllabus Tour
Encoding Structure: Part-based Models
Poselets (Bourdev et al)
Structureless Rigid
Deformable Part Models (Felzenszwalb et al)
Visual Recognition And Search
Check ICCV 2013 Tutorial on Partbased models for Recognition: http://www.cs.berkeley.edu/~rbg/ICC
V2013/
Columbia University, Spring 2014
Syllabus Tour
High-level Representation: Attributes and
Semantic Features
Output of Semantic Classifiers as Features, Zero-shot Learning, …
Check CVPR 2013 Tutorial on Attributes: https://filebox.ece.vt.edu/~parikh/attributes/
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Large-Scale Classification
Insights on how to handle huge amounts of data and categories
Deep Convolutional Neural Networks
Visual Recognition And Search
Check NIPS 2013 Tutorial on Deep Learning: http://cs.nyu.edu/~fergus/presentations/nips2013_final.pdf
Columbia University, Spring 2014
Syllabus Tour
Fast Indexing and Image Retrieval
Metric Learning
Indexing via Learning-based Hashing
Visual Recognition And Search
Database
Columbia University, Spring 2014
Syllabus Tour
Active Learning
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Case Studies
IBM Smart Vision Suite (SVS) IBM Multimedia Analysis and
Retrieval System (IMARS)
Visual Recognition And Search Columbia University, Spring 2014
Syllabus Tour
Basic Machine learning algorithms. We assume you are familiar with e.g., SVMs, …
Computer vision fundamentals. We assume you are familiar with e.g., edge detection, …
Visual Recognition And Search Columbia University, Spring 2014
Citizen Science Projects
Visual Recognition And Search Columbia University, Spring 2014
Cool Data for Projects
Snapshot Serengeti http://www.youtube.com/watch?v=AUL03ivS8bY http://www.snapshotserengeti.org/
Visual Recognition And Search Columbia University, Spring 2014
Cool Data for Projects
5 Terabytes of annotated data
Visual Recognition And Search Columbia University, Spring 2014
Cool Data for Projects
Example Project:
Build a web demo for fine-grained categorization of animals, using state-of-the-art visual learning techniques
Top classification results and associated confidence:
1. Zebra (0.98)
2. Striped Hyena (0.01)
3. Leopard (0.01)
We expect a comprehensive experimental analysis for each project
Visual Recognition And Search Columbia University, Spring 2014
Past 2013 Projects
Check the previous year project presentations
Visual Recognition And Search Columbia University, Spring 2014
See http://www.slideshare.net/cameraculture/raskar-ideahexagonapr2010
Visual Recognition And Search Columbia University, Spring 2014
Next Steps
Don’t forget to submit Homework #0 (Student Info
Form) if you are taking this class
Background in computer vision + coding skills in
Matlab or C/C++ are strict requirements for taking the course
If you don’t satisfy these requirements, please drop the class as soon as possible, as it will help us to know the final number of students in class.
Thank you!
Visual Recognition And Search Columbia University, Spring 2014