What is the mean of 'W_scale' in spatical_triplet_loss ?

[sheirving]

VideoTextSCM/train_embedding.py

Line 47 in d87ad1b

W_scale = torch.exp(1-torch.sum(torch.abs(triple_box[:, 0, 2:] - triple_box[:, 0, :2]), 1)/2)

Thankyou for share the code!
But what's the mean of 'W_scale' in spatical_triplet_loss, and loss_pos = pos_dist * (W_scale + W_pos_dis) - alpha1, why we need to add 'w_scale'?