Explore robust scene analysis techniques in this 47-minute conference talk by Stefan Roth at ROB 2018. Delve into challenges, reliability, and objectives in computer vision, focusing on robustness and accuracy. Examine disocclusion symmetry, KD 2015 results, and unsupervised learning for optical flow. Investigate domain mismatch issues and optical flow estimation methods, including mean field inference. Learn about motion segmentation, SiC Networks, and uncertainty sources in motion estimation and classification. Gain insights into full covariance and training objectives for improved scene analysis algorithms.
Overview
Syllabus
Introduction
Context
Challenges
Reliability
Objectives
Robustness
Accuracy
Dis occlusion symmetry
KD 2015 results
Optical flow unsupervised learning
Domain mismatch
Optical flow estimation
Optical flow training
Basic idea
Mean field inference
Takeaway
Motion Segmentation Example
SiC Networks
Uncertainty Sources
Motion Estimation
Optical flow
Uncertainty
Classification
Conclusion
Full covariance
Training objective
Taught by
Andreas Geiger
