In-Depth Computer Vision Research

The course will focus on a selected topic in computer vision that connects to the candidate’s doctoral thesis. The main aim of the course is to expand the research with intensive training on how to tease out the most relevant related works in computer vision, analytically or experimentally discover their drawbacks, make original contributions, and validate them. Computer vision involves the development of algorithms that can abstract complex unstructured data, such as images and videos, in a broad sense. This encompasses tasks like semantic segmentation, visual object tracking, object detection, extraction of 3D information, visual data generation, and their application to various downstream tasks like image/video manipulation and mobile robotics. The student will begin by selecting a suitable topic within their broader tentative doctoral research area, with the guidance of their supervisor. A specific goal, achievable within a single semester (approximately 150 hours of work), will be defined. The student will then establish a research plan and provide progress reports through bi-weekly meetings. The initial reports will involve breaking down and critically analyzing existing works closely related to the chosen topic from a methodological perspective. A series of test-and-hypothesis-generation sessions will be conducted, challenging the student to identify a localized big-picture idea that can lead to a scientific contribution in their selected area. This intensive training aims to equip the student with a robust methodology for approaching scientific discovery within computer vision and encourage them to generate their contributions. The ultimate objective is a high-caliber scientific contribution meeting publication quality standards necessary at a prominent computer vision venue. Moreover, the student will develop the ability to critically review relevant papers on arXiv as part of their daily routines (expected to spend an hour a day), ensuring they stay updated with daily scientific advancements. A secondary outcome of the course is to foster an understanding of the characteristics that distinguish top publications. The student will gain the knowledge and skills to reproduce the style and rigor required for high-quality scientific reporting in computer vision.

Restrictions/Prerequisites: The course is primarily aimed at doctoral candidates who have started their doctoral training under the mentorship of the course lecturer. The students' doctoral topic must be from the core computer vision topic. The appropriateness of the topic will be judged by the lecturer, and the candidate is invited to consult the lecturer before applying for the course.