Chinese Journal of Liquid Crystals and Displays, Volume. 40, Issue 8, 1219(2025)
Efficient siamese single object tracking based on hybrid feature fusion
Fig. 2. Hybrid feature fusion module. （a） Fast feature refinement unit； （b） Dual-branch feature aggregation unit.
Fig. 3. Success rates of FFTracker and other trackers on certain attributes
Fig. 4. Precision of FFTracker and other trackers on certain attributes
Fig. 5. Comparison of visual results of the algorithm proposed in this paper with two other efficient trackers on OTB100
Fig. 6. Comparison of visual results of the algorithm proposed in this paper with two other efficient trackers on LaSOT
Table 1. Experimental results on OTB100 dataset
View tableView in ArticleTable 1. Experimental results on OTB100 dataset
Backbone Module AUC/% P/% Parameters/M FLOPs/G Speed/fps EfficientNet-V1 ShuffleNet-V2 MobileNet -V3 MHA HFF 
42.6 57.3 5.7 0.5 194 
61.5 82.3 15.5 1.4 152 
59.4 80.5 13.5 1.2 144 
66.7 88.1 12.4 0.7 170 
57.3 77.3 15.2 1.4 135 
54.0 75.0 13.1 1.3 130 
63.0 86.4 11.6 0.8 141
Table 2. Experimental results for different stacking depths of HFF on the OTB100 dataset
View tableView in ArticleTable 2. Experimental results for different stacking depths of HFF on the OTB100 dataset
Parameters/M FLOPs/G AUC/% P/
%
GPU Speed/fps 4 18.4 1.0 64.6 87.5 92 3 15.7 0.9 66.1 88.0 90 2 12.4 0.7 66.7 88.1 170 1 9.1 0.6 64.2 85.6 227
Table 3. Performance comparison of the algorithm proposed in this paper with other object tracking algorithms
View tableView in ArticleTable 3. Performance comparison of the algorithm proposed in this paper with other object tracking algorithms
Trackers LaSOT OTB100 UAV123 AUC/% PNorm/% P/% AUC/% P/% AUC/% P/% Non-Lightweight ATOM［ 19 ］49.3 56.1 48.2 66.3 86.2 62.9 84.2 SiamGAT［ 20 ］53.9 63.3 53.0 71.0 91.6 64.6 84.3 DiMP50［ 21 ］56.5 65.0 56.9 67.6 87.3 65.0 85.6 SiamBAN［ 22 ］51.4 59.8 52.1 69.2 90.3 60.4 79.5 SiamRPN++［ 7 ］49.6 56.9 49.1 69.1 89.8 61.1 80.4 STARK-ST50［ 23 ］66.4 76.3 71.2 68.5 89.7 69.1 89.9 TransT［ 12 ］64.2 73.5 68.2 68.1 88.3 66.0 85.2 Lightweight LightTrack-M［ 18 ］52.5 58.6 51.7 66.1 86.0 46.9 60.9 HiT-Tiny［ 14 ］54.8 60.5 52.9 54.3 70.7 58.7 76.6 SiamFC［ 5 ］33.6 42.0 33.9 57.8 76.5 52.3 73.1 Ours 54.0 65.1 53.9 66.7 88.1 44.9 61.8
Table 4. Comparison of parameters， FLOPs and GPU speed on LaSOT dataset
View tableView in ArticleTable 4. Comparison of parameters， FLOPs and GPU speed on LaSOT dataset
Trackers Parameters/M FLOPs/G Speed/fps SiamBAN 53.9 48.8 40 SiamRPN++ 54.0 48.9 21 SiamFC 2.3 2.8 86 SiamGAT 14.2 17.3 70 DiMP50 43.1 10.4 57 ATOM 17.3 4.0 85 STARK-ST50 28.2 12.8 41 LightTrack-M 2.0 0.5 115 HiT-Tiny 9.6 2.0 73 TransT 23.0 16.7 84 Ours 12.4 0.7 170
Get Citation
Copy Citation Text
Na LI, Jinting PAN, Rongji LI, Yufei WANG. Efficient siamese single object tracking based on hybrid feature fusion[J]. Chinese Journal of Liquid Crystals and Displays, 2025, 40(8): 1219
Category:
Received: Apr. 28, 2025
Accepted: --
Published Online: Sep. 25, 2025
The Author Email: Na LI (lina114@xupt.edu.cn), Jinting PAN (lcdcpjt@163.com)
