Explore a comprehensive lecture on detector adaptation techniques presented by Enver Sangineto from the University of Central Florida. Delve into past and present research topics, focusing on the challenges of domain adaptation in pedestrian detection. Learn about innovative approaches, including a RANSAC-like method for target sample selection and spatial consensus collection. Examine training details, ensemble decision-making processes, and the implementation of spatially-dependent majority vote rules. Investigate various loss functions and RANSAC-like boosting techniques. Analyze the results and conclusions drawn from this cutting-edge research in statistical and spatial consensus collection for improved detector adaptation.
Statistical and Spatial Consensus Collection for Detector Adaption
University of Central Florida via YouTube
Overview
Syllabus
Intro
Past and present research topics
Detector adaptation
Outcome of a generic pedestrian detector
Domain Adaptation
A common approach
The proposed approach
Target sample selection
A RANSAC-like approach
Analogy with RANSAC
Training details
Collecting spatial consensus
Spatially-dependent majority vote rule for the ensemble decision
Spatial consensus algorithm
Initial classifier vocabulary cadinality
Ensemble cardinality
Simple majority vote
Different Loss functions
RANSAC-like boosting
Conclusions
Results
Taught by
UCF CRCV
