Computer Vision (6 ECTs)

Cameras are everywhere: phones, driving cars, hospitals, cinema, … even cleaning robots have one! But how do you go from raw pixel values into meaningful information? Where is the face in the image? Is there a pedestrian ahead? What is the 3D shape of the tumor in this lung? How do you add a virtual character into a real 3D scene?

This course provides a travel accros the field of Computer Vision: it starts at the basic concepts of light, color and camera devices and ends with cutting edge approaches recently published in the main scientific conferences of the domain. The course combines general technological basic concepts, theoretical mathematical fundations as well as practical sessions. In addition, as the Computer Vision field is in constant and rapid evolution, the course provides an emphasis on reading and understanding scientific papers.

Objectives

• Acquire knowledge on the basic concepts of vision and computer vision
• Acquire knowledge on the different strategies used for different tasks
• Acquire knowledge in the study of scientific papers

Lectures:

• Intro + Light color camera
• Filters contours segmentation + Interest points
• Deep Learning + features
• Projective Geometry
• Sigle Camera Model
• Panoramas + 2 view geometry
• Shape modeling: 3D perception, shape recovery basics
• Shape modeling: data-driven approaches

Scientific Paper Reading

• 1st Reading session on the plenoptic function: definition, acquisition, modeling and application.
• 2nd Reading session: datasets, optimization vs regression, neural representations, human body modeling applications.

Practical Units

• Basic camera calibration concepts with an AR Demo
• 3D multi-view reconstruction: voxel carving algorithm
• Neural 3D shape encoding with an MLP

Evaluation:

• Scientific Paper reviews and presentations (30%)
• Exam (70%)
• Practical Units are not graded - its content is evaluated in the exam

Lecturer: Sergi Pujades-Rocamora