Algorithm 1: Tracking with TFnet.
	1:	Input: Frames${\left\{I_i\right\}}_{i=1}^{N_f}$, initial target bounding box $b_1$
	2:	Augment the training samples according to the augmentation generation strategy.
	3:	Learn the variables in Equation (2) with the augmented training samples, where $s_k$ corresponding to the backbone network; t, the target template; $\varpi$, the foreground network
	4:	Copy the target template t to an adaptive template $t_a$ and static template $t_s$
	5:	For frame i = 2:$N_f$
	6:	Extract the search images according to the result in last frame $b_{i-1}$
	7:	Forward propagate the search images and predict the object location with Equation (4)
	8:	Generate a new training sample and its corresponding label based on the predicted       results
	9:	Update the adaptive target template $t_a$ with Equation (7)
	10:	End
	11:	Output: Tracking results ${\left\{b_i\right\}}_{i=2}^{N_f}$