This course is about information processing of knowledge, especially about computer vision. The classical methods of information acquisition, which are now often estimated by neural networks based on learning, will be studied, leading to a deeper understanding of events related to cameras, images, geometry, and reflective properties. Topics include image feature extraction, 3D shape reconstruction, object tracking, generative modeling, and reinforcement learning.
To be able to explain and implement basic topics about computer vision and reinforcement learning.
Computer vision, reinforcement learning, image features, object recognition, image segmentation, epipolar geometry, 3D restoration, optical flow, object tracking, image-based rendering, generative modeling
✔ Specialist skills | Intercultural skills | Communication skills | ✔ Critical thinking skills | ✔ Practical and/or problem-solving skills |
Lectures and programming exercises.
Course schedule | Required learning | |
---|---|---|
Class 1 | Neural networks | Topics include multi-layer perceptron, error backpropagation, convolutional neural networks, transformer, and network optimization. |
Class 2 | Image features | Topics include Harris corner detector, scale selection, SIFT, HOG, ORB, and object detection. |
Class 3 | Object recognition, metric learning, image segmentation | Topics include face detection, identify recognition, pedestrian detection, gestalt principle, K-means/Mean-shift clustering, graphcut, superpixels. |
Class 4 | Epipolar geometry, 3D reconstruction | Topics include epipolar geometry, essential matrix, fundamental matrix, robust estimation, Structure from Motion (SfM). |
Class 5 | Optical flow, object tracking | Topics include optical flow, aperture problem, tracking, particile filter、data association. |
Class 6 | Special imaging devices, generative models | Topics include plenoptic function, light field, high-dynamic range imaging, neural rendering, generative models. |
Class 7 | Reinforcement learning | Topics include Markov decision process, optimal action value function, Q learning. |
To enhance effective learning, students are encouraged to spend approximately 100 minutes preparing for class and another 100 minutes reviewing class content afterwards (including assignments) for each class.
None.
Computer Vision: Algorithms and Applications
The level of understanding and ability to apply the lecture content will be evaluated by submitting reports on the assignments.
No course requirements.
reikawa[at]sc.e.titech.ac.jp