Interested in computer vision, I took COMP6528 Computer Vision during my first semester at ANU also due to my previous work in this field. I’m grateful to our lecturers, Dr. Miaomiao Liu and Dr. Dylan Campbell, and the tutors for their dedicated efforts and guidance. I have strengthened my ability in computer vision through this course and I highly recommend COMP6528 to every student studying computing here.
COMP6528 covers low-level, mid-level, and high-level vision, as well as 3D vision. During the first six weeks before the mid-break, Dr. Miaomiao Liu taught us about low-level vision and most of the mid-level and high-level vision topics. After the mid-break, Dr. Dylan Campbell took over and focused on 3D vision. In the last two weeks, we shifted our focus back to mid-level and high-level vision topics like optical flow, shape from X, detection, and segmentation.
You might be wondering what low-level, mid-level, and high-level visions are? Low-level vision includes image formation, image representation and processing, and image filtering. In the mid-level part, Dr. Miaomiao Liu led us through edge detection and image features, including classic corner detection algorithms like Harris and SIFT. For high-level vision, the course covered the principles of deep neural networks, including multilayer neural networks, backpropagation algorithms, activation functions, convolutional networks, pooling layers, batch normalization layers, and loss functions.
Personally speaking, the most valuable part of the course was the 3D vision section since it was the first time for me to learn 3D vision systematically. I have acquired 2D vision during my undergraduate and worked with it as an algorithm engineer in computer vision. Unlike traditional computer vision courses that focus only on image processing and deep learning, the 3D vision section in this course was particularly beneficial. It covered topics like camera models, homography, camera calibration, and two-view geometry (epipolar geometry, triangulation, and stereo cameras). These 3D vision concepts are essential components of a 3D reconstruction system.
The COMP6528 course in 24S1 includes three assignments, accounting for a total of 45 marks, 15 mark for each. Each assignment assesses the knowledge learned at the current stage, focusing on concept understanding and code implementation. If necessary, I will upload my assignment reports on GitHub for those who are interested can follow and share.
The final exam accounts for 55 marks. And the 3D vision is the most significant part of the course, accounting for 50% of marks in the final exam. The questions of 3D vision in the final exams are often presented in the form of calculation questions and comprehensive questions.