Lecture Notes
Slides
Slides below may not be exactly the same as those presented in class. In case of substantial changes,
the updated slides will be reposted below.
Video lectures are posted on Micorosft Teams as the course progresses.
Such posted videos are not the recordings of the live lecture/seminar sessions taking place on Tuesdays and Thursdays.
The posted videos explain all the slides in details, including review of basic backgrounnd in linear algebra and multi-variate calculus.
While the live lectures/seminars aim at covering all the required material, they are more focused on the main points
specific to computer vision and provision time for answering students' questions.
For some points not covered in class, the students could be referred to the posted video lectures.
- Topic 1: Introduction
- Topic 2: Image modalities, pin hole camera model.
- Topic 3: Low-level features: image filtering, convolution, cross correlation, gradients, edges, corners, Gaussian pyramid,
detectors+descriptors.
- Topic 4: Image warping (domain transforms): linear, affine, homography. Forward and inverse warping.
- Topic 5: Panoramas, projections and homographies, image blending
- Topic 6: Geometric model fitting: matching, least squares, RANSAC, multi-model fitting.
- Topic 7: Multi-view geometry: projection matrix, camera calibration, epipolar geometry, essential/fundamental matrices, structure from motion.
- Topic 8: Dense stereo: rectification, disparity, depth map, window-based stereo, scan-line stereo, regularization over grid.
- Topic 9 A, B: Low-level segmentation: unsupervised and interactive, region- and boundary-based, objectives (criteria, losses, regularization), K-means, EM, kernel and spectral clustering, mean-shift, graphical models, graph cut, variance and entropy clustering.
- Topic 10: Classification: supervised machine learning basics (from linear to non-linear classifiers), training losses, gradient descent, CNNs, class-activation maps (CAM).
- Topic 11: Semantic segmentation: fully convolutional networks, encoder/decoder, detection, transfer learning.
- Topic 12: Weak and Self-supervision: regularized losses, monocular depth, NeRF, inpainting, generative models.
- Topic 13 (bonus): Deep clustering, self-augmentation, self-labeling, collision cross-entropy