Tracking Persons-of-Interest via Adaptive Discriminative Features

The 8 challenging music videos tested in our experiments are publicly available on YouTube. In Table 1, we list the links of all music videos. The sequences T-ARA, WESTLIFE, and PUSSYCAT DOLLS are live music concert recordings and acquired from multiple cameras with different views. The other sequences BRUNO MARS, APINK, HELLO BUBBLE, DARLING, and GIRLS ALOUD are MTV videos taken in different scenes. All music videos contain large face appearance variations across different shots due to changes in pose, view angle, scale, makeup, illumination, camera motion, and heavy occlusion. In Figures 1–4, we show randomly selected sample faces in temporal order in the videos for each person (using ground truth annotations) to illustrate the intra-class variations and inter-class variations on four challenging sequences (APINK, DARLING, T-ARA and BRUNO MARS). Table 2 summarizes the statistics of these videos, including the duration, frames, and the number of shots, tracklets, detections, and main casts.