Overview
This course focuses on exploring label-efficient visual abstractions for autonomous driving. The learning outcomes include understanding the impact of reducing annotation costs on behavior cloning agents and analyzing segmentation-based modalities for self-driving tasks. The course teaches practical insights on exploiting visual abstractions in a label-efficient manner to enhance driving performance. The teaching method involves a keynote presentation with a focus on theoretical concepts and practical applications. The intended audience for this course includes researchers, practitioners, and enthusiasts in the fields of computer vision, autonomous driving, and machine learning.
Syllabus
Introduction
Two dominating paradigms to selfdriving
Direct perception
Conditional forints learning
Intermediate representations
More Related Findings
What is a good visual abstraction
InputOutput
No Crash Benchmark
Identifying Relevant Classes
Results
Qualitative Results
Summary
Dataset Overview
Illustrations
Taught by
Andreas Geiger