CS088/CS188 | Web-powered computer vision

Course description

The last few years have seen a dramatic growth of photo-related Web sites, such as picture/video sharing services and image search engines. For example, it is estimated that several billion images are currently indexed and searched by Google and Bing Images, while over three billion personal photos have been uploaded to Flickr. Many of these images come with associated text or tags loosely describing them. From a computer vision point of view, these sites represent extraordinarily rich and diverse repositories of weakly-labeled image examples that can be exploited for object recognition, image understanding and 3D modeling. Conversely, computer vision techniques can be applied to design new tools allowing users to browse and search photos in these large databases more effectively. Examples of such applications include image-based search of products, automatic photo annotation and geo-localization, video summarization, and 3D navigation of pictures.

The objective of this course is to expose students to this fascinating emerging field of research through paper reading and discussion. Students will take turns presenting articles and leading the in-class discussion. All students will be required to read the papers presented in class before the lecture and to submit written critiques addressing the main contributions and weaknesses of the work described in the articles. In-class participation will count toward the final grade. Students will also be required to propose and complete a term project. There is no midterm or final exam.

Administrative information

Instructor: Lorenzo Torresani | 109 Sudikoff | office hour: T-Th 4:00-5:00pm
Teaching assistant: Xiaochao Yang | 111 Sudikoff | office hour: T-Th 4:00-5:00pm
Lectures: T-Th 2-3:50pm | x-hour (used sporadically for lectures) W 4:15-5:05
109 Dartmouth Hall; Note that the first lecture will be on Tuesday September 29, and not on Thursday September 24.

Grading and policies

Grading scheme

The final course grade will be based 10% on in-class participation, 20% on the written critiques, 20% on the paper presentations, and 50% on the term project.

Prerequisites

Discussion of the papers should focus on the high-level concepts and on the application ideas rather than on the implementation specifics. Thus, there are no formal technical prerequisites besides the ability to understand the high-level ideas in a vision paper. However, prior exposure to computer vision and machine learning will be useful (although not required) for the project. You obviously must be familiar with a programming language (e.g. C++, Matlab) in order to complete the term project. If you are not sure you have the right background to take this class, please come talk to me.

Late submissions

Late submissions will not be accepted under any circumstances: you will get a zero grade for any late submission. However, you have three free critiques: you are allowed not to submit critiques for up to three of the papers that we will be discussing in class. This is done to allow you to cope with unexpected circumstances, such as sickness. After you have used up your three free critiques, you will penalized for further failing to submit critiques.

Auditing

Please contact the instructor if you would like to audit the course.

Academic integrity

In order to encourage independent critiquing and to foster in-class discussion, you will NOT be permitted to talk about the assigned papers with your classmates before the lecture: I am interested in hearing your own personal views of the articles and not consensus opinions that would emerge from group discussions held before the lecture.

You are allowed to use external software for your project. However, you should clearly report the use of external code and include pointers to such software in your project write-up. The project grade will be based on the novelty of your solution/application but also on the amount of new code written by you to implement the idea. So keep this in mind when considering to use software written by someone else.

These rules will be strictly enforced and any violation will be treated seriously