ComputerVision

This GitHub repository contains the assignments and project work for a Computer Vision course, covering topics such as image formation, 3D voxel data construction, image features, optical flow, image classification, and convolutional neural networks (CNNs).

Folder Structure:

• Assignment 1: Image Calibration

• Assignment 2: Voxel Based Reconstruction

• Assignment 3: Color Model and Voxel Labeling

• Assignment 4: CNN Model Training and Validation

• Assignment 5: Action Recognition

• Outputs : collection of output images

Assignment 1 involves image calibration, while Assignment 2 focuses on 3D voxel-based reconstruction with a task description including the quality of intrinsics, extrinsics, and background subtraction. Assignment 3 builds on the previous work and requires the creation of a color model for each person in a video and labeling voxels based on this color model. Assignment 4 covers CNN model training and validation, and Assignment 5 focuses on action recognition in still images and videos using transfer learning, optical flow, and CNN outputs.

Course Discribtion

The goal of computer vision is recognize and understand the world through visual information such as images or videos.

After completing the course, the student: understands the motivation and goal of computer vision, including the applications and general challenges. understands the mechanisms of image formation in terms of both geometry and radiometry. understands and is able to construct 3D voxel data from images or videos based on silhouettes. understands and is able to cluster data points based on various distance measures. understands the concepts of image features and their importance in computer vision. understands the concepts and challenges of optical flow. understands the challenges of image image classification and object detection. understands and able to express the performance of classification and detection algorithms. understands the components and the learning mechanisms of convolutional neural networks (CNN). understands the training of CNNs, and is able to develop and evaluate CNNs for image and video classification tasks.
Assessment

This course is about the algorithms and mechanisms to extract and classify information from images and video. The course combines theory and practice, with two themes: multi-view reconstruction and CNN image/video classification.

Course ID

Utrecht University INFOMCV https://osiris.uu.nl/osiris_student_uuprd/OnderwijsCatalogusSelect.do?selectie=cursus&collegejaar=2021&cursus=INFOMCV

Programming Languages

C++, Python

Team

Vlad Kalyuzhnyy (Vlad-2299) and Idan Grady