Question about losses in you paper and code

[massyzs]

Hi there, it's relative question that has been closed by others. #24
Related with ranking loss:

1.The way to select neighbor of label0 and label1 here. Is it to set a window and select two sub-windows on two opposite corners?

2. folds Can you explain why using folds here and will this be kind of hyperparameters?

3. Is this the object position in the image? If so, that means you have a coarse segmentation or grounding?

Smooth Loss: Could you please tell me which part is for this loss?

Appreciate your answer~