hinge loss question

[albertofernandezvillan]

In the paper, they employ a modified hinge loss function:
loss[i, j] = max(1 − α · (D[i, j] − T[i, j]) · B[i, j], 0)

Here, the training procedure uses hinge as loss function:

Train:
  optimizer: adam
  loss: hinge
  ....

So I understand that the weights included here (Sauvola_demo.h5) are obtained using hinge loss, while the official pre-trained weights (Sauvola_v3_att_w7.15.23.31.39.47.55.63_k1_R1_a1_inorm_S256_R0.02_B8_lr0.001_E30-Acc0.9895-Tacc0.9282-F0.9448-PSNR20.59.h5) are obtained using modified hinge loss function.

Is this correct?