|
-
Apr. 14. The final grades (and also the marks for the final exam) have been published on BrightSpace. If you any questions, please contact me before April. 23rd.
-
Apr. 7. More details about the final exam is online here. Good luck!
-
All news ...
|
About
Photogrammetry and 3D Computer Vision (i.e., 3DV) aim at recovering the structure of real-world objects/scenes from visual data (i.e., images, point clouds). This course is about the theories, methodologies, and techniques of 3D computer vision for the built environment. In the term of this course, you will learn the basic knowledge and algorithms in 3D computer vision through a series of lectures, reading materials, and lab exercises.
Contents
The topics of this course cover the whole pipeline of reconstructing 3D models from images, as well as semantic segmentation/classification of point clouds:
-
Cameras models: how a point from the real world gets projected onto the image plane and how to recover the camera parameters from a set of observations;
-
Image matching: define image features to establish correspondences between images, i.e., finding the same points in image pairs;
-
Structure from motion: recover/refine geometry and camera parameters simultaneously from image correspondences;
-
Multi-view stereo: recover dense geometry (e.g., point clouds) from images;
-
Surface reconstruction: obtain 3D surface models of real-world objects from point clouds;
-
Semantic segmentation: classify/segment the reconstructed 3D point; into semantically meaningful object classes.
Goals
After finishing this course, you will have built a working knowledge of the theory, methodology, and algorithms/techniques used in 3D computer vision. Specifically, you will be able to:
-
apply linear algebra knowledge to analyze and evaluate computer vision algorithms;
-
explain and analyze the core algorithms in 3D computer vision (e.g., camera models, image matching, structure from motion, multi-view stereo, and surface reconstruction);
-
choose appropriate methods for reconstructing smooth surfaces and piecewise planar objects;
-
propose and implement solutions for reconstructing real-world buildings from images;
-
extend and improve the existing piecewise planar object reconstruction algorithms;
-
apply state-of-the-art machine learning methods for semantic segmentation/classification of point clouds.
Assessment
The assessment of this course consists of 3 group assignments and the final exam. The final grade will be based on the contribution in doing the assignments and the final exam, breaking down to:
-
group assignments (40%). The assignments are assessed based on your first submission. If you improve and resubmit your assignment based on feedback (from the teachers or peer-reviewing), you may receive a slightly higher mark, but no more than G0 + 0.5 (G0 denoting the mark of the first submission);
-
final exam (60%). The final exam consists of multiple-choice questions and open questions. Example questions will be given two weeks before the exam.
Both assignments and the final exam have to be graded sufficiently (5.5 minimum) to pass the course. A total of 6.0 or above is necessary to successfully pass the course.
Communication
Based on past experiences of teaching staffs and students preferences, we will mainly use the following communication tools for our course:
-
Zoom for lectures. Lecture recording will be made available after on BrightSpace (for unlimited and permanent storage).
-
Discord channels for general notification and Q&A.
In addition to the regular course hours, it is also possible to meet the teachers individually (by appointment only). When you make an appointment, please indicate two time slots and a link to the meeting room (could be Zoom, Teams, whatever).
Lecturers and teaching assistants
Copyright @Liangliang Nan. 2021
