COMPSCI 670: Computer Vision (Grade: A)

Course Overview:
This graduate-level course delves into the sophisticated techniques for analyzing visual data, particularly color images. The course is structured in two key parts:

1. Image Formation and Representation:
   • This section explores the physics and geometry behind image formation, including camera design and color sensing in the human eye. We examined topics such as radiometry, light, color, and image formation principles that are fundamental to understanding how images are captured and represented digitally.
2. Algorithms and Applications:
   • The second half of the course shifts focus to algorithms that extract meaningful information from images. This includes:
     • Image Alignment and Depth Estimation: Detecting interest points and aligning images for accurate depth analysis.
     • Object Detection and Segmentation: Learning both classical methods and modern, learning-based techniques for detecting and segmenting objects within images.
     • Generative Modeling: Understanding the architecture and applications of models that can generate new images from learned data distributions.

Coursework and Assignments:
Throughout the course, I engaged with a series of assignments that emphasized key vision tasks and methods. Each assignment involved constructing a basic vision system, followed by iterative improvements through error analysis and model redesign. The final project allowed for an in-depth investigation of a specific topic or application, enabling the application of learned concepts to real-world challenges.

Key Topics Covered:

• Radiometry and Light: Understanding the interaction of light with surfaces and how it is captured by cameras.
• Image Formation: Exploring the geometric and optical principles behind image capture.
• Linear Filtering and Image Processing: Techniques for modeling and manipulating images.
• Optical Flow and Feature Matching: Methods for analyzing motion and matching features across images.
• Neural Networks for Vision Tasks: Applying deep learning models to tasks like image classification and object detection.
• Transfer Learning and Advanced Recognition: Leveraging pre-trained models for specialized tasks in computer vision.
• Generative Adversarial Networks (GANs): Creating new images through adversarial learning.
• Unsupervised Learning and 3D Shape Understanding: Techniques for learning from unlabeled data and understanding three-dimensional shapes.

Achievement:
I completed this challenging course with a perfect score of A (4.0/4.0), reflecting my deep understanding and capability in the field of computer vision.

Instructor:
The course was taught by Grant Van Horn, a renowned expert in the field, who provided invaluable insights into both classical and modern techniques in computer vision.