Fig 2 Field Enhancement Refinement.
Focus-MOT has four loss functions, which correspond to the loss hm_loss
for heatmap, wh_loss for boxsize, off_loss for offset, and id_loss
for Re-ID. for hm_loss, we use the MSE loss function to calculate; for
wh_loss, we use the L1 loss function to calculate; for off_loss, we
use the multivariate cross-entropy function to calculate. use the L1
loss loss function to calculate; for off_loss, we use the L1 loss
function to calculate, and for Re-ID loss we use the multivariate cross
entropy function to calculate.
Then the total loss is:
\begin{equation}
\text{loss}\ =\ \text{hm}\_\text{loss}\ +\ \text{wh}\_\text{loss}\ +\ \text{off}\_\text{loss}\ +\ 0.1\times\text{id}\_\text{loss}\nonumber \\
\end{equation}Focus-MOT uses the training set provided by the six datasets MOT17,
Caltech, Citypersons, Cuhksysu, PRW, and ETH, and the test set of MOT15
and MOT17 is used for testing [8-12].