Fig. 1. (a) Two-stream two-stage model with light weight relational modeling; (b) Two-stream two-stage with heavy relation modeling; (c) One-stream one-stage without extra relation

Fig. 2. (a) The structure of the proposed algorithm; (b) Low-rank side network; (c) Spatial feature enhancement module

Fig. 3. Visualization of the attention weights of search region corresponding to the center part of template after different Transformer layers

Fig. 4. (a) The forward and backward process of the network in fully fine-tune; (b) The forward and backward process of the network in this method

Fig. 5. The precision rate，normalized precision rate and success rate curve of various tracking algorithms in LSOTB-TIR-120. (a) Precision rate curves; (b) Normalized precision rate curves; (c) Success rate curve

Fig. 6. The precision rate and success rate curve of various tracking algorithms in PTB-TIR. (a) Precision rate curve; (b) Success rate curve

Fig. 7. Visualization of tracking results of different video sequences

Fig. 8. Tracking performance and GPU peak occupancy under various M. (a) Tracking precision and GPU peak; (b) Tracking norm precision and GPU peak occupancy; (c) Tracking success rate and GPU peak occupancy