Exploiting the Circulant Structure of Trackingbydetection with Kernels
Exploiting the Circulant Structure of Tracking-by-detection with Kernels Seunghoon Hong CV Lab. POSTECH
Motivation • Tracking-by-detection – A classifier is trained in on-line using examples (patches) obtained during tracking
Motivation • Sampling strategy – Sparse sampling for computational efficiency
Motivation • Redundancy in sampling – A set of dense patches has extreme redundancies
Motivation • Key idea – Representing a set of dense samples by “circulant structure” – Training and testing can be extremely efficient by exploiting such structure
Circulant matrix •
Circulant matrix •
Circulant matrix • Efficiently computed by Fast Fourier transform !
Circulant matrix •
Circulant matrix • shifting direction
Learning with dense sampling •
Learning with dense sampling • Kernel matrix
Learning with dense sampling •
Learning with dense sampling •
Learning with dense sampling •
Learning with dense sampling •
Learning with dense sampling •
Learning with dense sampling •
Learning with dense sampling •
Overall algorithm
Experiments • Original image After preprocessing
Experiments • Training label – Smooth regression output by Gaussian kernel
Experiments
Q&A
- Slides: 24