Kursens utveckling

A full revision of the lectures was made this year. Every slide was updated, with some material being either discarded or replaced. Material related to image enhancement and segmentation was considerably shortened and replaced by two new lectures, an introduction to deep learning and a new lecture on novel view synthesis. The goal was to focus on traditional methods of computer vision that are just as relevant today, as they were before deep learning was introduced, as well as on topics today dominated by deep learning. Another significant change in the course was the introduction of an online quiz for grade E in Zoom. There were two motivations for such a change, to promote continuous learning throughout the course and encourage students to complete the voluntary quizzes during the course and to relieve students from some of the burden they experience when preparing for the exam by allowing them to focus on more problem-related questions.

Due to the pandemic, the course was run in a hybrid format with lectures on campus but also broadcast in Zoom. Lab presentations and help sessions were however held fully online, but this year there was an option to do the labs in Python instead of Matlab. Q&A sessions were introduced, during which students could ask any question related to the course. The course content was updated for about half the lectures with more examples of deep learning-based methods for computer vision in particular.

Due to the pandemic, the course this year had to be changed into a fully online format, where only the exam was on campus. Students that could not physically attend the exam were given the opportunity to attend the re-exam that was also held online. Lectures were given in Zoom and recorded with videos published on Canvas. Lab presentations and help sessions were also held online. However, while presentation sessions were densely scheduled during the last days of each lab week, help sessions were more sparsely spread over the whole period giving students and TAs more flexibility.

A change compared to earlier course rounds introduced this year, was to more actively use Canvas to provide direct feedback and help, in particular for questions related to the labs. TAs were staffed to moderate discussions related to different aspects of the course and provide quick feedback to students in need of help. This encouraged students to directly ask for help, rather than wait until the dedicated lab help sessions. Another change was that the re-exam was held online due to the COVID pandemic. With questions posted online in Canvas students worked from home while being connected through Zoom, but without the necessity of having the camera turned on. Canvas was also used to check IDs, while the chat in Zoom could be used for questions.

The course was updated with more examples of deep learning being used in computer vision, in order to better align with courses in machine learning. A short introduction to typical neural networks in computer vision was given in the first lecture, which is followed by examples in various lectures about feature matching, segmentation, object recognition and motion analysis. A set of voluntary quizzes were also introduced after each lecture. The quizzes serve two purposes. First, it gives students an idea of how much in depth they need to go into each individual topic. Since the course is introductory it spans the whole field of computer vision and the amount of available literature is vast. A second reason is to focus on important concepts that are often misunderstood. Quiz questions are often phrased such that incorrect answers easily lead to cognitive dissonance when an explanation is given at the end. Finally, the quiz gives the lecturer feedback on what should be reiterated during lectures.

Besides some updates of the course material itself, a few important changes were made to the lab course to reduce the burden on teaching assistants and make students focus more on the essentials: 1) Students were asked to upload to Canvas summaries of lab results with reflections on key concepts from the labs. 2) Students presented their labs during 20 minute individual sessions, instead of 30 minute sessions that were used before. This was possible by focusing on key concepts from the labs, rather than on the results, results that had already covered in the uploaded reports. 3) Instead of using computer halls, a regular seminar hall was used. Rather then showing results on-screen, the uploaded reports were used as support for the students in their presentations. The goal of these changes were to force students to reflect more on key concepts and less on the code and results.