Seminar Introduction
"Vision is a process that produces from images of the external world a description that is useful to the viewer and not cluttered with irrelevant information." (Marr and Nishihara).
In the past few years, significant progress has been made in the development of new technologies and in their use as the core of several real-time vision systems for detecting specific classes of objects, including people, faces and cars, within complex images. In our seminar we will explore several of the more successful image representations and supervised learning algorithms that enable computers to start tackling the challenging task of image understanding.
Seminar Topics
We will cover a variety of object recognition topics including:
- Vision as an inverse problem: The different types of object recognition problems. Pose and illumination variations and invariant representations. The problem of clutter in images.
- Local image representations: A comprehensive study of modern image representations such as SIFT, C2, HOG, gray level patches, etc.
- Statistical models: Generative and discriminative models. Basic graphical models. Off-the-shelf discriminative classifiers (SVM, boosting). Avoiding over-fitting and learning from few training examples.
- Learned image representations: Classical and Modern dimensionality reduction algorithms; Basic feature selection techniques.
- Combined image segmentation and recognition: The connection between the problems and proposed solutions, including: Jigsaw models, backprojection-based, OBJCUT.
- Video analysis: From images to video; representing motion; combining detection with tracking.
We will examine many real-world applications such as: face identification, vehicle detection and behavior analysis.
Seminar Schedule
We shall meet every Wednesday 14:00-16:00. Detailed schedule available here. I have asked Rotem (rotem.littman@gmail.com) to assist with assigning weeks to students.
Class Presentations
Please send me your presentations after you present them in class and correct them, (1) as a pdf file (you may want to use PDFcreator) and (2) as .jpg file of the first slide (size should be 307x230 pixels, you may want to use irfanview). Please make sure to include in each slide that you borrow explicit credit to the creator of that slide. Please put this credit on the lower right corner.
Contact Me
If you are interested in learning more about computer vision or this seminar feel free to contact me.
Basic Details
- Instructor: Dr. Lior Wolf
- Meetings: Wed 14-16
- Location: Kaplun 324
- Office hours: Mon 14-16
- Course ID: 0368-4606-01
- Prerequisites: none
- Important note: If you are interested in object recognition but were unable to take the seminar, you are welcome to contact me.
Useful Terms
- Object Recognition: Object Detection + Object Identification.
- Object Detection: The recognition of a learned object class. Example: finding cars in a street scene.
- Object Identification: An individual instance of an object is recognized. Example: identification of a specific person face.
- Image Classification: Image detection applied to the whole image. Example: "is this an image of a tiger?".
Example Tasks
Annotation of street images
