USES is keras, loss of dice_loss, two classification problem of image segmentation, why run label=1 label ious in the process of falling, finally reduced to 0 don't move?

 def dice_loss (y_true y_pred, smooth=1) : 
Y_true_f=K.f latten (y_true) 
Y_pred_f=K.f latten (y_pred) 
Intersection computes=Keith um (y_true_f * y_pred_f) 
Loss +=(2 * intersection computes + smooth)/(Keith um (y_true_f) + Keith um (y_pred_f) + smooth) 
1 - return loss 

The strange thing is, when I change the code to the

 def dice_loss (y_true y_pred, smooth=1) : 
Y_true_f=K.f latten (y_true [...] 1) 
Y_pred_f=K.f latten (y_pred [...] 1) 
Intersection computes=Keith um (y_true_f * y_pred_f) 
Loss +=(2 * intersection computes + smooth)/(Keith um (y_true_f) + Keith um (y_pred_f) + smooth) 
1 - return loss 

Iou index will rise in normal again, this time, there are bosses know what's the difference between the two calculation formula? The second code can represent binary classification dice_loss?