SKT-MOT and DyTracker: A Multiobject Tracking Dataset and a Dynamic Tracker for Speed Skating Video

Table 6

Results of applying KFDU and FDU to various MOT methods.

Method	MOTA	IDF1	MT	ML	FP	FN	IDSW
DeepSORT + KFDU + FDU	76.80 (+0.1)	77.9 (+0.6)	56.7	11.8	3,165	9,910	231
FairMOT + KFDU + FDU	69.1	73.4 (+0.6)	41.3	15.6	2,011	14,480	283
ByteTrack + KFDU	76.6 (+0.1)	79.7 (+0.2)	59.6	11.8	3,431	9,680	243

The TBD methods use the YOLOX detector [31]. All experiments are performed on the MOT17 validation set.