2020/2021 - 4th Year, 2nd Semester
Course: Visão por Computador (VCOM) | Computer Vision Authors: Bernardo Santos(bernas670), David Silva (daviddias99), Laura Majer (m-ajer) Luís Cunha (luispcunha)
Description: In this work we explore some common tasks of modern Computer Vision. We apply classic machine learning algorithms (SVMs) with bag-of-words descriptors and state-of-the-art deep learning architectures to a multi-class classification problem of objects in a Museum (the Met). Since the dataset is highly unbalanced dataset, we explore techniques such as data augmentation. The architectures we use are then adapted to use in a multi-label classification problem. Finally we explore the usage of CAMs (class activation mapping) to tackle a painting object detection task.
For more information on the specification see docs/specification.pdf and for a detailed report on our work see docs/report.pdf.
Technologies: Python, OpenCV, Jupyter Notebooks, Scikit Learn, Tensorflow, Keras
Skills: Object detection, artwork classificiation (single and multi-label), bag-of-words, SVM, CNNs, data augmentation, CAM (class activation maps), machine learning, deep learning
Grade: 19.3/20