Algorithm 2: DCF-DSST Tracker: Iteration at Frame t

Inputs:     $I_t$: Image patch     $p_{t-1}$: previous frame target position     $s_{t-1}$: previous frame target scale     $y_{trans}$: regression target for translation model (Gaussian shaped)     $y_{scale}$: regression target for scale model (Gaussian shaped)  Outputs:      $p_t$: detected target position     $s_t$: detected target scale  Training: 1 Compute the Gaussian kernel correlation between x and itself, $k^{xx}$, using (A2) 2 Compute the DFT of the solution coefficients in the dual space, $A_{trans}$, using (A1)  Translation Detection: 3 Construct the test sample, ${\mathbf{z}}_{t,trans}$, from $I_t$ at $p_{t-1}$ and $s_{t-1}$ 4 Compute the correlation response, $f\left({\mathbf{z}}_{t,trans}\right)$, using (A3) 5 Maximize the response, $f\left({\mathbf{z}}_{t,trans}\right)$, to find target position, $p_t$  Scale Detection: 6 Construct the test sample, ${\mathbf{z}}_{t,scale}$, from $I_t$ at $p_t$ and $s_{t-1}$ 7 Compute the correlation response, $f\left({\mathbf{z}}_{t,scale}\right)$, using (A10) 8 Maximize the response, $f\left({\mathbf{z}}_{t,scale}\right)$, to find target scale, $s_t$  Update: 9 Extract training samples ${\mathbf{x}}_{t,trans}$ and ${\mathbf{x}}_{t,scale}$ from $I_t$ at $p_t$ and $s_t$ 10 Update the translation model using (A4) and (A5) 11 Update the scale model using (A7) and (A8)