Recent Advances in Computer Vision

CSE 291 Topics in Computer Vision, Winter 2018

Instructor: Manmohan Chandraker
Email: mkchandraker [AT] eng [DOT] ucsd [DOT] edu

TA: Shashank Shastry
Email: scshastr [AT] eng [DOT] ucsd [DOT] edu

Lectures: WF 5-6:20pm at CSE 4140
Instructor office hours: F 3-4pm at CSE 4122
TA office hours: Tu noon-1pm at CSE B215

Overview

The advent of deep learning has led to significant gains in computer vision. In this course, we will study advances in several areas of computer vision, such as image classification, object detection, semantic segmentation, 3D reconstruction and activity recognition. We will study these by discussing a mix of canonical and recent publications. A focus area will also be to study weakly supervised, self-supervised and domain adaptation frameworks in computer vision.

Prerequisites

This is an advanced class, covering recent developments in computer vision research. Students at all levels including undergraduates, masters and PhD, with a strong interest in computer vision may enroll. Comfort with optimization, linear algebra, probability and statistics is necessary. Prior background in computer vision and machine learning is desirable, preferably through research experience or as covered by CSE 152, 252 and similar offerings. Contact the instructor if unsure about meeting criteria for enrollment.

Course Format and Requirements

The course will involve presenting and discussing recent papers related to computer vision. The papers will be grouped by concept. These will be supported with lectures by the instructor to provide historical background or introduce fundamentals. Students will be required to do a project or write a critical analysis on a focused topic. Wide flexibility is available for the choice of topic, but should be discussed with the instructor beforehand. The best projects may consist of conceptual development that pushes the boundaries of the state-of-art, or a practical implementation that goes beyond the ones associated with the papers read in class.

Students may take the class with a letter grade option, which requires one or two in-class presentations (depending on enrollment strength), critical reviews, a final exam and completion of the project. Alternately, a 1 unit, S-U or P-NP registration is available, which requires in-class presentations and reviews. You may also audit, but a registration is preferred.

Grades will be weighted as 20% for in-class presentations, 20% for reviews, 20% for final exam, 10% for class particiation and 30% for the project. The project may be in groups of three. The goal of the course is to develop an understanding of the current state of computer vision and gain appreciation of its limits and potential.

Topics

The course will cover a diverse range of topics in computer vision. An inexhaustive list includes:

Outline

Jan 10: Lecture Jan 12: Lecture Jan 17: Lecture Jan 19: Lecture Jan 24: Correspondence and stereo Jan 26: Optical flow Jan 31: Depth estimation Feb 02: Face recognition Face alignment Human pose estimation Material and illumination Object detection Object detection Image captioning Multi-target tracking Action recognition Semantic segmentation Unsupervised domain adaptation Generative Adversarial Networks

Resources

Manmohan Chandraker
Last modified: Tue, Mar 20, 2018