Fig. 1. Framework of proposed algorithm

Fig. 2. Architecture of the SK attention network

Fig. 3. Architecture of the MSI module

Fig. 4. Visualization results of originalimage and features from corresponding layer.(a) Original image centered on the target; (b) low level features; (c) high level features; (d) fused version generated by MFB

Fig. 5. Architecture of the multi-scale channel attention network

Fig. 6. Comparison of success rate and precision on OTB100 dataset. (a) Success rate; (b) precision

Fig. 7. EAO performance on VOT2018 dataset

Fig. 8. EAO under different visual attributes on VOT2018

Fig. 9. Comparison of success rate and accuracy on LaSOT dataset.(a) Success rate; (b) normalized precision

Fig. 10. Comparison of speed and performance of different algorithms on LaSOT dataset

Fig. 11. Tracking results of five algorithms